Factual Recall

Reasoning and agentic workloads increasingly demand efficient long-context inference. Yet full-attention decoding reads the growing history at every step, regardless of its benefit to the next prediction. We show that a pretrained model's decoding states already contain information predictive of this benefit, before the global read. Building on this finding, we introduce On-Demand Attention (ODA), a local-first decoding method that uses a lightweight recall head to selectively invoke global attention as its predicted benefit changes during generation. ODA trains only the recall head, leaving pretrained weights unchanged and the complete historical KV cache available for future recall. We further implement GPU-side conditional execution in vLLM, translating reduced global reads into practical decoding speedups over full attention at long context lengths. Experiments across Qwen and Gemma models, including hybrid-attention backbones, show that selective recall recovers most of the performance lost under local attention while substantially reducing global reads. These findings support long-context inference in which pretrained models guide their own access to the information they retain.

Clinical coding agents repeatedly encounter the same failure modes, including unsupported codes, missed documented conditions, specificity errors, and procedure-coding convention mismatches. We introduce Learn-Then-Act, an inference-time adaptation framework that converts errors from a small labeled LEARN batch into a structured Mistake Knowledge Database (MistakeKDB). False-negative lessons are routed to a recall-oriented Coder, while false-positive lessons are routed to a precision-oriented Judge. We instantiate the framework in LearnActCoder, a Coder-Judge clinical coding pipeline with lookup-table grounding where available. On 150 matched MIMIC-III notes, structured MistakeKDB improves CPT F1 by 5.9 percentage points, while raw-example and reflection-style memories remain near the no-memory baseline; the ICD-9 improvement is not significant. On a matched MIMIC-IV cohort, memory shifts ICD-10 coding toward higher precision at a recall cost, leaving F1 statistically unchanged. Applying the same memory to 1,000 held-out MIMIC-III notes maintains a stable ICD operating point, providing scale/stability evidence. Overall, the results are consistent with structured, feedback-derived error memory being useful for adapting clinical coding behavior across cases without weight updates or changes to the underlying workflow. Absolute CPT/HCPCS performance remains low, and the system is evaluated retrospectively rather than in clinical deployment.

We describe our system for LitTraceQA (GroundLM @ EMNLP 2026): given a research question, retrieve the relevant papers from a pool of 27,487, cite the page and the table or figure where the answer lives, and answer in a requested format. Our main finding is that evidence grounding is limited by retrieval, not by reading. The page selector put the annotator's page, which we call the gold page, in front of the model that locates evidence only about half the time (52.6% gold-page recall), while that model, given the page, cited the right one in 45 of the 48 locators it emitted (94%). When the page was missing it rarely said so: of 45 such cases it returned nothing 14 times, a wrong page 24 times, and a correct page 7 times, so the pipeline failed quietly almost twice as often as it failed visibly. Since the failure was that the right page was never shown, the fix is to stop choosing: each retrieved paper fits in the model's context, so we show it whole. Page ranking survives only as a fallback inside papers too long to fit, which no test-split paper was, and gold-page recall reaches 100% on the papers we can parse. Separately, questions that identify their target by position rather than content, such as "the first author of the 24th reference", are served by parsing rather than retrieval: we resolve the bibliography into an addressable list, which also supplies identifiers the evidence metric scores. The final system scores 0.762 paper $F_1$, 0.441 evidence $F_1$ and 0.920 multiple-choice accuracy on the held-out test split. Because the pipeline depends on a closed model without seed control, we release a harness that verifies the paper's central claims against committed artifacts.

Many operational cases are documented more than once, at different workflow stages and for different purposes, yet model evaluations normally select one of these records before model comparison begins. We treat that selection as part of the evaluation and compare matched records of the same cases under fixed labels and splits in three systems: GE Aerospace repair events, NASA ASRS safety reports and NHTSA vehicle recalls. Across the three GE fields, for events whose label comes from parts transactions independently of the narratives, held-out macro-F1 ranged from 0.33 to 0.91. A difference of 0.46 separated the customer report, written before shop work, from the technician report, written after diagnosis but before the transaction that generates the label. That difference is substantially larger than the representation and architecture differences tested on the same events. The public systems showed different patterns: the NHTSA defect summary remained strongest under every model family tested, whereas the ASRS analyst synopsis outperformed the reporter narrative under learned sequence models but not under lexical baselines. Secondary analyses showed that some model comparisons were also record-dependent. Evaluations should be run on the information available at the intended decision point and should report how both the record and the label were produced.

Factual hallucination in closed-book question answering is often treated as a coverage problem: a model fails because the relevant fact is absent from its internal memory. This view misses a second source of error. Even when a fact has been observed, finite memory may force it to be stored only approximately. We study this effect through a simple coverage--compression model of factual recall. We consider an unstructured question-answering task with $N$ possible queries and $K$ possible answers. A learner observes $M$ training facts, compresses them into at most $B$ bits, and answers uniformly drawn test queries without retrieval. For a uniformly random ground-truth mapping, we prove $\mathcal{E} \geq \frac{M}{N}\delta^\star\!\left(\frac{B}{M}\right) + \left(1-\frac{M}{N}\right)\left(1-\frac{1}{K}\right)$, where $\delta^\star(r)$ is the inverse rate-distortion function of a uniform $K$-ary source under zero-one loss. The two terms separate compression distortion on observed facts from missing coverage on unobserved facts. The bound gives a compact way to reason about selective memory, forced compression, structure, retrieval, abstention, and long-context organization. We study the predicted signatures with theory-implied simulations and controlled fact-injection probes in modern language models that vary fact load and effective trainable memory. The result is not a complete theory of hallucination, but an information-theoretic account of a separable failure mode: lossy recall of observed facts under finite memory.

Open-world object detection (OWOD) extends closed-set detection by requiring models to identify unknown objects and incrementally learn them once annotations become available. In remote sensing imagery, object categories often exhibit latent hierarchical relationships that may be inadequately represented in the Euclidean spaces commonly adopted by existing methods, limiting unknown-object recall and incremental-learning performance. To address this issue, we investigate hyperbolic geometry for OWOD in remote sensing imagery and propose HyRS-OWOD. To improve unknown object recall, we design a two-step unknown-object discovery mechanism: a Decoupled Objectness Learning (DOL) module that disentangles foreground perception from semantic information to separate foreground proposals from background regions, followed by a Hyperbolic Uncertainty Learning (HUL) component that leverages the radius of hyperbolic embeddings as an uncertainty-aware cue for known-unknown discrimination. For incremental learning, we develop a Hyperbolic Metric Learning (HML) strategy that enhances inter-class separability, facilitating the incorporation of novel categories while mitigating catastrophic forgetting. Experiments on three remote sensing benchmarks demonstrate consistent improvements in unknown recall and incremental learning over state-of-the-art OWOD methods.

Accurate citations are the foundation of academic writing, tracing intellectual origins and substantiating core claims. However, manually navigating the growing volume of scientific literature is increasingly difficult, prompting reliance on automatic citation recommendation. While modern retrieval-augmented architectures have largely mitigated the fabrication of non-existent papers, current systems relying on semantic similarity struggle with misattribution, often citing authentic papers that fail to logically support the author's claim. To address this challenge, we argue that accurate citation requires a shift from similarity-based search to active, claim-level reasoning. We propose ReCite, a decoupled agentic framework that orchestrates location perception, intent-aware query planning, and reflective verification. Trained on synthesized reasoning trajectories, our agent verifies claim-evidence consistency and triggers self-correction loops when retrieved candidates lack logical support. Experiments demonstrate that our lightweight framework outperforms state-of-the-art massive generative models in strict citation accuracy. By grounding literature matching in verifiable logic rather than semantic overlap, ReCite establishes a reliable foundation for automated academic writing.

Updating a language model's knowledge through fine-tuning is essential for keeping its outputs current, yet can also induce factual forgetting and new hallucinations. Prior work shows that long-tail knowledge is harder to acquire and newly memorized long-tail facts are difficult to retain during later fine-tuning. We study a complementary question: among facts that a model has encoded correctly, which are most vulnerable to collateral corruption during other updates? To investigate this question under a realistic factual distribution, we construct a large-scale graph FACTPROP of verified Wikipedia facts by linking triples that share head or tail entities, thereby preserving connections among factual knowledge. We fine-tune models on factual statements and measure correct-to-incorrect facts after each update. Our results reveal a pattern distinct from prior findings on long-tail vulnerability during acquisition and retention: among facts that models already answer correctly, those associated with highly connected entities are more likely to be corrupted by neighboring updates, and updates to such facts propagate errors more broadly. Structural popularity therefore predicts both vulnerability and downstream damage. Inspired by this finding, we propose Popularity-based Anchoring (PopAnchor), a lightweight rehearsal strategy that preserves a small set of popular facts and reduces forgetting.

Associative Recall (AR) is the cognitive ability to learn and retrieve links between items in memory. In NLP, AR is used as a benchmark for evaluating the in-context memory capacity of architectures such as Mamba, and has been found to strongly correlate with language modeling performance. This paper explores AR from the perspective of mechanistic interpretability, aiming to reverse-engineer the exact internal algorithm used by Mamba to perform recall. Our key insight is that Mamba performs recall by implicitly learning linear hash functions, and we identify the low-level circuit that enables this behavior. Building on these findings and inspired by theoretical tools in similarity-preserving hashing, such as the Johnson-Lindenstrauss lemma, we develop a theoretical framework for analyzing AR, which we term Recall Scaling Laws. Given the vocabulary size and the number of facts in context, this framework allows us to (1) predict the embedding and state dimensions required for Mamba to achieve perfect recall, (2) predict recall success probability given the model dimensions, and (3) analyze multi-layer models and multi-head SSM patterns. Empirical results show that our theoretical findings are accurate and predictive, offering insights into how AR capacity scales with vocabulary, state, embedding size, and architecture.

Gaps remain in our understanding of how large language models (LLMs) acquire knowledge during pre-training. We posit that auxiliary views, reformulations of knowledge, are causally helpful for learning. We design controlled experiments to isolate this. First, we confirm that repetition is necessary for acquisition and clarify that paraphrasing helps only at smaller batch sizes. Second, holding the token budget fixed, allocating tokens from document repetition to auxiliary views improves learning, counterintuitively, even for factual recall. Third, the effectiveness of auxiliary views is not contingent on the strength of the teacher model that generates them. Fourth, we identify forms of knowledge, contextual and foundational, that aid learning in the presence of prior knowledge gaps. Finally, we examine how these effects manifest mechanistically via layer-wise biases and compression. Together, our findings suggest that auxiliary representations of knowledge, which arise naturally in large pre-training corpora, are a key factor in the success of pre-training and offer a plausible explanation for why data diversity matters.

LLM-empowered paper-code discrepancy detection has received growing concern since the scaling of research submissions exceeds the manual review capability. However, the limited context capacity and one-sided discrepancy detection of existing single-agent LLM paradigms lead to an inferior recall performance in detecting discrepancies. In this paper, we propose Dude, the first Dual-Detection Multi-Agent System for paper-code discrepancy detection. We discover that the granularity asymmetry of the paper-language and code-language introduces over-interpretation and over-reporting challenges in a multi-agent system design for discrepancy detection, resulting in increasing false positives. To address this, we propose a granularity-aligned negotiation and a two-stage salience-filtering mechanism in Dude, which effectively prevents agents from falsely reporting discrepancies. Experimental results in real-world paper-code discrepancy datasets showcase Dude's significant recall and precision improvement by up to 22.8%, increasing F1 score by up to 18.7% compared to baseline methods.

Reliable attractor recall conventionally requires broad basins of attraction. However, in reservoir-computing based associative memory, temporal cues reliably recover dynamical memories despite basins dominated by unpredictable, riddled-like regions. We reveal that memory basins exhibit an octopus-like'' structure: a robust head'' near the attractor and thin, intertwined ``tentacles'' spanning state space. Initial states in tentacular regions yield near-zero uncertainty exponents, making the recalled memory effectively unpredictable at finite precision. Yet, cue-driven generalized synchronization bypasses this unpredictability, driving the system into the robust basin head. This mechanism yields a quantitative relation linking minimum cue duration, synchronization rate, and basin-head radius. Trained recurrent neural networks exhibit similar geometry, suggesting this phenomenon extends beyond reservoir computing.

Logit-based knowledge distillation (KD) is used to train smaller language models (LMs) via supervision from stronger teachers, but whether its benefits are consistent across training stages remains unclear. Through controlled experiments, we find that forward Kullback-Leibler (KL) distillation--the standard KD formulation--with post-trained teachers behaves fundamentally differently during mid-training, an intermediate phase of self-supervised learning on curated corpora. Surprisingly, while forward KD simultaneously improves reasoning and factual recall during pre-training relative to standard next-token prediction (NTP), it instead slows factual recall acquisition during mid-training despite continued reasoning gains. We trace this stage dependence to an asymmetry in teacher confidence across data domains and the student's evolving knowledge state: teachers are more confident on procedural than knowledge-intensive data, while students acquire low-entropy factual knowledge earlier in training. To mitigate this imbalance, we propose Switch Distillation, a simple mid-training objective that distills on tokens where the teacher is confident, using teacher predictive entropy as a lightweight routing signal, and otherwise falls back to cross-entropy. Switch Distillation consistently outperforms existing distillation objectives across teacher sizes. Relative to standard NTP, it achieves 1.61-1.71x the reasoning performance and 1.13-1.19x the knowledge and commonsense performance while preserving 96.7-96.8% of factual recall. Crucially, these benefits persist after post-training: Switch Distillation closes the factual recall gap while maintaining 1.25-1.32x and 1.13-1.20x gains in reasoning and knowledge and commonsense, respectively.

We investigate how vision-language models (VLMs) handle context-memory conflicts; that is, situations in which the model is given information in context that differs from what was stored parametrically during training. We document asymmetric biases: models tend to prefer in-context information about entities which appear in text, but prefer parametric information about entities which appear in images. We relate this asymmetry to the late representational alignment across modalities, showing that the longer processing time associated with resolving visual entities prevents the suppression of the model's usual factual recall mechanism, thus resulting in more parametric answers. Chain-of-thought reasoning does not appear to resolve the gap, but increasing the amount of visual information in the context does show an effect. These results illustrate the complexity of ensuring consistent behavior as models become increasingly multimodal and retrieval-augmented.

Long-term memory is increasingly important for conversational agents, yet existing benchmarks primarily measure memory through pointwise factual recall: whether a system can recover isolated facts or event-level details from prior interactions. Real-world memory use, however, often requires a more demanding capability: integrating distributed, implicit, and noisy evidence across extended interaction histories into coherent, task-oriented outputs. We call this capability memory utilization. Here, we introduce UtilMem, a diagnostic benchmark comprising 1,717 instances across five domains, designed to evaluate four underexplored aspects of memory utilization: reasoning over dense histories, identifying implicitly relevant memories, synthesizing distributed evidence into summaries, analyses, or plans, and resisting interference from semantically similar distractors. Evaluating a diverse set of retrieval-based and memory-augmented systems, we find that strong performance on conventional factual-memory benchmarks does not reliably translate into effective memory utilization. Moreover, retrieval alone is insufficient: even when relevant evidence is successfully recovered, systems frequently fail to integrate information across sessions or to distinguish useful evidence from plausible distractors. These findings expose a substantial gap between accessing stored information and using it effectively, and suggest that progress in long-term conversational memory will require architectures that explicitly support evidence integration and robustness to retrieval interference. Code is available at https://github.com/peijunallin/UtilMem.

Recurrent, attention-free sequence models share a structural weakness: a fading state cannot perform exact recall of something seen once, far in the past. We add to the Kathleen trunk a second memory layer -- a "notebook": a fixed-key holographic (HRR) associative store with a learned local write gate, a self-gating raw read, and write-triggered forgetting -- 25K parameters that attach to the logits of any trunk. (1) Mechanism: on a controlled needle-in-haystack task the notebook reaches 80-82% one-shot recall at 4x the training length, where the bare trunk scores ~4% and a parameter-matched attention head scores 100% inside its training length and 0% beyond it. Addressing is length-invariant by construction; the untrained memory alone recalls at 90% accuracy identically at 512, 2048 and 4096 bytes. Because the store is a linear superposition, two capabilities follow from arithmetic alone: selective unlearning (one subtraction erases one fact to chance, retained facts unharmed) and per-token attribution (counterfactual erasure names the source fact of every correct byte, 100% provenance). (2) Real text: on WikiText-2 bytes the notebook improves prediction of repeated rare words by +0.15-0.27 bits/byte, the gain growing with the distance between mentions and holding zero-shot at 4x training length; write-triggered forgetting eliminates memory pollution at 8x length (first-mention cost +0.33 -> -0.004). (3) Scope and scale: a parameter-matched attention head does generalize on natural-text repetition, so the notebook's claim is exact recall at O(L); on a WikiText-103 ladder (8 to 512 MB) the zero-shot repeat gain rises monotonically. All experiments are pre-registered, seeds reported, and reproducible on a single free-tier GPU.

Vision-Language Models (VLMs) are highly effective in retrieving semantically relevant images. However, in practice, relevance alone is often insufficient. Systems must also achieve Result Diversification (RD) across composite attributes such as geography and time, a task for which precise control remains challenging. Current re-ranking methods, such as Multi-Source Determinantal Point Processes (MS-DPP), address this using manifold-based repulsion over similarity representations. Although this strategy is effective for broad exploration, it exposes a key limitation in manifold-based models: when subjected to diversity-decrease tasks on discrete metadata, they suffer substantial degradation in early-rank recall. To bridge this gap, we introduce MASCOT (Model-Aware Submodular Coverage for Composite-Attribute Text-to-Image Retrieval). Instead of relying on manifold repulsion, MASCOT formulates multi-attribute diversity as a resource allocation problem, projecting attributes into a soft-binning space weighted by query-driven importance. Averaged across the three PixelProse diversity-decrease tasks, MASCOT preserves an early-rank recall (R@10) of 88.58%, while MS-DPP retains 67.63%. The margin widens under composite constraints: on PP_geo_hour, where temporal and geographic diversity must be suppressed simultaneously, MS-DPP's recall collapses from 0.9737 to 0.4931 and its top-ranked result degrades to R@1 = 0.23, while MASCOT holds R@10 = 0.9410 and R@1 = 0.7202 at a diversity metric above the unconstrained baseline. We do not claim uniform superiority: on aggregate diversity-relevance scores our own simpler ablations attain higher harmonic means on all three decrease tasks, and MASCOT's advantage is specific to recall beyond rank 1 under composite constraints.

Long-running conversational agents increasingly rely on a memory system to avoid resending the whole conversation each turn, yet how much that costs to serve has received little systematic benchmarking. We compare three memory systems (Mem0, Hindsight, and Mastra Observational Memory) against two reference strategies -- a fixed-size rolling window and resubmitting the full transcript -- across two backbones and conversations of up to 400 turns, pairing every cost measurement with answer accuracy on 665 LoCoMo questions. First, a memory system's serving cost cannot be predicted from conversation length and message size alone: a regression that tracks the two reference strategies closely misses the memory systems by 18-69%, their cost driven instead by internal memory behavior. Second, a break-even analysis shows that whether -- and when -- a memory system becomes cheaper to serve than the full transcript is highly sensitive to the system and the backbone, from the first tens of turns for the cheapest to never within 400 turns for the most expensive. Third, no system wins on both axes: accuracy spans 21-54%, and the backbone choice drives cost as much as the memory system does.

Generalized Hopfield networks are introduced where memories and neurons are continuous variables that lie on a Riemannian manifold. We explicitly focus on symmetric spaces associated with the special unitary groups SU(d), and use both numerical and analytical (replica) techniques to demonstrate an almost order of magnitude enhancement in critical capacity over the vector networks starting with d=3 and further rapidly growing with d. To circumvent the non-linear geometric constraints, we use a Lie algebraic method [following V. Galitski, Phys. Rev. A 84, 012118 (2011)] to exactly describe the classical neural network in terms of linear algebra in an auxiliary Hilbert space. It is shown that in contrast to the traditional Hopfield networks, memory recall in SU(d) Hopfields corresponds to neuron alignment along a top eigenvector of a spiked matrix, which is less susceptible to random matrix crosstalk than other models with continuous neuron variables. Physical platforms to realize SU(d) Hopfields are briefly discussed and physical (in addition to algorithmic) recall mechanism is demonstrated, where memory recovery occurs naturally through generalized Landau-Lifshitz-Gilbert dynamics. To illustrate SU(3) memory recall, we introduce a color (RGB) image encoding/decoding protocol and explicitly run image recovery on corrupted cues. Finally, we quantize the generalized Hopfields which are shown to reduce to Sachdev-Ye glassy type of models. Their many-body spectra generally feature two types of dark and memory bands, where the latter exhibits chaotic Wigner-Dyson level statistics that hides Hebbian data.

Visual retrieval-augmented generation (RAG) commonly expands the retrieved evidence set to improve answer-page coverage, implicitly assuming that all available evidence should be passed to the generator. We show that this assumption does not hold for diffusion language models (DLMs): retrieving more pages increases answer-page recall, whereas unconditionally passing all retrieved pages to the generator often reduces answer accuracy, primarily because of semantic conflict. A latent-source analysis explains this mismatch through source-coherence loss in parallel denoising, where position-wise proposals can combine incompatible visual sources into unsupported answers. We further find that such interference is already visible in the first-step answer-block distribution, making it possible to assess evidence before decoding. To preserve retrieval coverage while limiting harmful visual exposure, we propose the Entropy-Based Candidate Filter (ECF), a training-free evidence-admission framework. To reduce irrelevant content within individual candidates, ECF constructs multi-granularity evidence units; to identify beneficial additional evidence, it uses blank-controlled block confidence and retrieval rank to determine whether and which candidate should enter the final context. Across three multimodal DLMs and five visual QA benchmarks, ECF improves answer accuracy by 2.62 percentage points on average over the strongest fixed top-$k$ input and, with LLaDA2.0-Uni, by 2.37 percentage points on average over the best competing training-free result for each dataset. These results show that broader retrieval benefits visual DLM-RAG through selective evidence admission rather than unconditional evidence expansion. Code is publicly available at https://github.com/wjkuser/ECF.

Scaling supervision for multi-turn medical agents is difficult because expert dialogue annotation is costly and clinical conversations are privacy-restricted. We introduce Guideline-as-Oracle (GAO), which compiles American Academy of Ophthalmology guidance into a 70-row operational rule table and uses it as the sole source of instance-level supervision for 3,000 training dialogues, reserving human labeling for evaluation. Because converting rules into dialogues is itself a design problem, we catalog eight construction strategies, including cited-row tier assignment, one-fact boundary pairs, metadata-only repair, and label repair, and characterize the evidential status of each: labeling mechanism, null, confounded, or evaluated only as a package. Fine-tuning a 9B backbone on this corpus yields GAO-Triage, improving agreement with a 201-case operational reference from 61.7% to 74.1% (exact McNemar p=0.0046) and emergent-case recall from 9.5% to 69.0%; the gains persist across a second seed and patient simulator. None of the seven general-purpose systems we test dominates GAO-Triage on both metrics, and GAO-Triage requires no frontier model at inference time. Permuting label-dialogue assignments collapses the model to a constant-routine predictor, indicating that the signal lies in guideline-derived assignment rather than dialogue surface form. Label repair coincides with the disappearance of a late-training safety degradation.

The Rashomon effect describes the phenomenon that many models can achieve nearly equivalent performance on the same learning task, with significant ramifications for robustness, feature importance, and customizability. These use cases motivate the computation of Rashomon sets: the set of all models whose regularized loss is near-optimal. Decision trees are one of the few model classes for which Rashomon sets can be fully enumerated, but this computation has always been conditional on a binarization of the original data, either restricting which splits each tree is allowed to make or substantially increasing the complexity of an already difficult combinatorial problem. We introduce the first algorithm that exactly enumerates decision-tree Rashomon sets while exploiting the ordered structure of continuous features. We further develop a relaxation for approximate enumeration and an anytime algorithm that progressively refines the set of candidate thresholds, producing increasingly detailed approximations that converge to the continuous-feature Rashomon set. Experiments show that coarse binarization can miss many trees, important features, and predictive multiplicity; our algorithms achieve orders-of-magnitude speedups over existing enumeration methods, with approximations providing further speedups while maintaining near-perfect recall.

Large language model (LLM) agents are increasingly used as personalized assistants in high-stakes domains such as financial advising, yet it remains unclear whether they can maintain and update an individualized user model over long horizons. Existing personalized-memory benchmarks primarily test factual retention or rely on weakly constrained model-generated trajectories, leaving event-driven preference adaptation underexplored. We introduce FinPerMA, an event-grounded benchmark that evaluates personalized memory against frozen longitudinal investor trajectories. Its generation pipeline combines deterministic, theory-informed impact rules, controlled LLM narration, and automated quality screening; a Post-Shock checkpoint isolates whether an agent has integrated a material event into its persistent user model. On 2,994 questions from 276 personas, seven frontier LLMs and up to seven memory configurations remain far from saturated: no full-context configuration exceeds approximately 0.47 overall accuracy or approximately 39% on multiple-choice questions. Attribution analysis shows that summary-based memory often preserves factual details while losing the preference signals needed for personalization; simple retrieval can therefore outperform purpose-built memory systems, with the gap widening after shocks.

A wearable assistant should both answer questions about its visual history and recognize when that history is useful to the present situation. Existing video-memory systems primarily support question-conditioned recall, whereas proactive assistants typically use separate memory and control mechanisms. We introduce GROVE, a training-free framework that supports both behaviors with one memory grown causally from a continuous video stream. GROVE retains fine-grained perceptual evidence and incrementally consolidates it into time-stamped moments, coherent episodes, and recurring cross-day patterns. Each stratum is paired with a scale-native retrieval skill for locating an observation, replaying an activity, or traversing long-range regularities. Reactive QA and proactive assistance share this memory and access interface, differing in whether retrieval is initiated by a user query or the current situation. Across multiple benchmarks including the challenging MM-lifelong and EgoServe, GROVE achieves the best results among the compared methods. Controlled ablations show that the temporal strata and their access skills are complementary, with patterns providing the largest benefit when evidence spans multiple days. Code will be available at https://github.com/SitongGong/GROVE.

Long-term memory is critical for LLM agents operating over long-horizon interactions. However, several persistent limitations of existing memory systems can be traced to two recurring misalignment patterns in long-term interaction settings: Temporal-Structural Misalignment (TSM) and Delayed Utility Manifestation (DUM). TSM arises when temporal proximity does not reliably align with topical or event-level relatedness, whereas DUM arises when write-time salience does not reliably predict future query utility. To mitigate these misalignment patterns, we propose MemSIF (Memory with Structured Interactions and Facts), a structured interaction-to-fact memory framework. Structured Interaction Memory organizes raw interactions into Topical Segments that preserve local topical coherence and Event Trajectories that maintain cross-time event continuity. Dual-Track Fact Memory uses two complementary tracks: CoreFact memory consolidates stable, schema-guided information at write time, whereas ActiveFact memory forms facts on demand and promotes those supported by multiple historical sources and recurring query demand for reuse. Experiments on LoCoMo and LongMemEval-S across five backbone LLMs show that MemSIF achieves the highest Total ACC in all settings, outperforming the strongest baseline by 2.29%-8.79% on LoCoMo and 2.87%-6.15% on LongMemEval-S. These results support the effectiveness of combining Structured Interaction Memory with Dual-Track Fact Memory to mitigate TSM and DUM. Code is available at https://github.com/luoyufeihaha/MemSIF.

Lexical retrieval (BM25) captures exact keyword matches and weights terms by corpus-wide significance, but it is blind to the semantic vocabulary gap: when a relevant document phrases an answer differently from the query, BM25 never retrieves it, and no amount of downstream reranking or fusion can recover a document that was never in the candidate set. We present Cross-Encoder Query Expansion (CE-QE), which reads the per-token relevance attributions of a cross-encoder applied to top semantic search results, selects the terms the cross-encoder treats as decisive, and appends them to the BM25 query. Unlike classical pseudo-relevance feedback, which reuses BM25's own (possibly wrong) top results, CE-QE seeds expansion from the semantic retriever's results, avoiding self-reinforcing query drift. Unlike recent generative query expansion (HyDE, Query2doc), which prompts a large language model to hallucinate text from its parametric knowledge, every CE-QE expansion term is copied verbatim from a retrieved passage, so it cannot introduce vocabulary the corpus does not contain, and its only added cost is attribution extraction on a cross-encoder a hybrid pipeline already runs for reranking. On seven BEIR datasets, CE-QE improves lexical recall substantially where query and answer vocabulary diverge (e.g., NQ Recall@100 from 0.32 to 0.47), and its score-fusion variant (SESF) beats cross-encoder score fusion by 2.5% on Recall@100 and beats SPLADEv2 and ColBERTv2 by 5.3% and 4.6% on nDCG@10, while leaving the underlying BM25 index completely unmodified.

Top-$K$ sparse attention reduces the cost of Softmax and value aggregation by attending to only a small subset of key--value (KV) entries. However, identifying this subset still requires scoring the current query against the full KV cache and performing global Top-$K$ selection, leaving selector cost linear in context length and limiting the practical efficiency of sparse attention for long-context decoding. In this paper, we introduce ReTopK, a training-free method that accelerates dynamic Top-$K$ attention by reusing historical retrieval decisions. ReTopK builds on the observation that similar queries often attend to overlapping supports and that partially overlapping supports can still preserve most of the Exact Top-$K$ attention mass. For each attention head, it maintains a bounded cache of historical query--support pairs, retrieves the most similar cached queries for each new query, unions their stored supports with a recent window, and reranks only the resulting compact candidate set using exact current-query scores. A similarity-based fallback invokes full-history Exact Top-$K$ when reuse is unreliable, while periodic exact refreshes limit cache drift. ReTopK retains the complete KV cache and reuses only selected indices, rather than historical scores, attention weights, or outputs. Across 16K--128K contexts, ReTopK achieves the lowest PG19 perplexity and the highest NIAH and LongBench scores among the evaluated approximate methods. At 128K with $K=512$, ReTopK incurs only a 0.50% perplexity increase over Exact Top-$K$ while accelerating attention computation by $3.07\times$.

Large language models are increasingly deployed with persistent personalized context, such as accumulated memory profiles or long conversation histories, that is shared across a user's many requests. Production memory systems (e.g., Mem0, MemGPT, and Zep) retrieve a relevant subset of this memory and inject it into the prompt, forcing the serving engine to repeatedly prefill the same content. As the retrieval budget grows, time-to-first-token (TTFT) increases even though the underlying memory is reused across requests. We present InferScale, a GPU-native LLM memory system that replaces repeated prompt prefilling with reusable KV state. InferScale precomputes each memory fact's KV representation, stores it alongside a semantic embedding on the GPU, retrieves relevant facts at serving time, and injects their KV directly into vLLM's paged cache. To support dynamically assembled memories under rotary position embeddings, we introduce Chunked RoPE, which stores keys before rotation and applies their serving-time positions during injection. However, encoding memory facts independently omits the cross-fact context available during joint prefilling. We mitigate this with Context-Window Encoding, which encodes each memory fact together with a small window of preceding conversation context while caching only the target fact's KV. InferScale is implemented through vLLM's KV-connector interface, requiring neither engine modifications nor model fine-tuning. Across three open-weight models on LoCoMo, InferScale keeps TTFT nearly constant as the retrieval budget increases: at k=50 it reduces TTFT by 72-79% (3.6-4.8x), achieves 60.3% accuracy versus 63.3% for Mem0 without serving-time recomputation, and delivers 3.7-4.5x the throughput under concurrent load. Reusable KV state thus decouples memory-conditioned serving latency from retrieved-context size while preserving application quality.

Dual-hand action segmentation commonly fuses left- and right-hand representations at identical temporal indices, although coordinated hand transitions may occur with nonzero and time-varying delays. We introduce Lag-Aware Cross-Hand Alignment (LACA), a lightweight module that explicitly estimates directional temporal-offset distributions between hand-specific feature streams. LACA retrieves cross-hand information from the estimated offsets and incorporates a learned null state to suppress transfer when no compatible cross-hand transition is supported. Alignment is supervised using compatibility-aware targets derived automatically from frame-level training annotations, without requiring additional labels. Analysis of the HA-ViD and ATTACH training annotations reveals robust nonzero cross-hand matches for 44.7% and 48.9% of transition anchors, respectively, compared with 18.6% and 21.3% under temporally shifted controls. When integrated into Polyphony, LACA improves the two-hand mean F1@50 from 40.4 to 42.5 and boundary F1 from 56.5 to 59.6 on HA-ViD, and from 19.9 to 21.8 and 44.7 to 47.9, respectively, on ATTACH, relative to our reproduced Polyphony baseline. These gains require only approximately 0.0086 million additional trainable parameters. We further introduce LACA-C, a future-free variant that restricts alignment and the complete inference pipeline to current and past observations. On ATTACH, LACA-C achieves 83.6% transition-cue recall, a seed-averaged median availability delay of 233~ms, 0.72 false cues per minute, and segmentation-stage throughput of 224.9 current-position predictions per second. These results demonstrate that explicit cross-hand temporal alignment improves both action segmentation and boundary localization while supporting timely future-free perception.

Long-horizon LLM agents accumulate reasoning traces, actions, and tool observations that can eventually exceed a model's fixed context window. Existing compaction methods address this limitation by discarding, summarizing, or retrieving earlier information, but they may remove task-critical details or fail to recover them reliably. We propose ARC (Addressable Recall Compaction), a context-management framework that separates archival storage from active-context presentation. ARC stores tool observations in an append-only, ID-addressable log and replaces older observations with compact citations when compaction is required. The agent can subsequently use these identifiers to request stored content without re-executing the corresponding tools or depending solely on similarity-based retrieval. We evaluate ARC using Qwen3-8B with a 16k context window and Qwen3-32B with a 32k context window. On the Needle-in-a-Haystack evaluation, ARC achieves an average exact-answer accuracy of 99.40%, compared with 88.12% for the best-performing baseline in our evaluation. ARC also reduces estimated serving time and HBM traffic under our hardware-cost model. On the LongBench-v2 Hard subset, ARC obtains an average accuracy of 29.97%, compared with 28.25% for the best-performing baseline. These results indicate that explicit, address-based recall can improve information retention and serving efficiency relative to the evaluated context-management baselines under the tested settings.

Memoir combines per-sample fast memory, shared slow parameters, variable-depth latent recurrence, and a future-latent energy objective. We test its riskiest coupling: each pondering iteration may rewrite the fast tier that the same iteration reads. On procedural associative recall with key interference, we compare a coupled arm against an otherwise identical read-only pondering arm. Both arms contain 81,738 parameters, including 76,362 trainable parameters, and use matched declared forward multiply-accumulate counts, data, optimizer, schedule, and seeds. After 240 training steps across 12 seeds, coupled recall is 0.5203 with a 95 percent interval of [0.4522, 0.5883], while read-only recall is 0.6557 with [0.5953, 0.7160]. The arms are paired per seed, and the read-only lead of 0.1354 gives a paired t of 3.23 on 11 degrees of freedom with a 95 percent interval of [0.0431, 0.2277] on the difference, winning on 10 of 12 seeds. After 960 steps across 8 seeds, both arms reach 1.0000, so the measured effect is a learning-speed penalty at a fixed budget, not a demonstrated capability penalty. That longer control is ceiling limited, leaving convergence on a non-saturating task unmeasured. A predicted failure in which memory rewriting corrupts the energy signal did not occur: the energy margin grew and held. Kernel restructuring also reduced delta-rule forward time from 0.907 ms to 0.351 ms on the stated device. Code and evidence are available at https://github.com/RightNow-AI/Memoir

Language models encode substantial factual knowledge in their parameters, which can lead to unreliable behavior when this knowledge is outdated, incomplete, or misaligned with the provided context. In this work, we study whether modifying the pretraining signal can systematically shift models away from parametric recall and toward evidence-grounded reasoning. We introduce Knowledge--''Less'' Language Models (KLLMs), a fundamentally different epistemic training paradigm for LLMs, which are pretrained on corpora in which named entities are anonymized, thereby removing a primary channel for entity-linked factual supervision. This intervention substantially reduces closed-book factual recall, while often improving performance on tasks where relevant information is provided as context. Across multiple model scales, KLLMs consistently outperform matched baselines on contextual question answering, fact verification, and hallucination detection benchmarks. Crucially, in retrieval-grounded settings with imperfect evidence, KLLMs show improved robustness and achieve up to 20--25% relative gains over standard language models. They further exhibit better calibration, with improved ECE, Brier score, and AUROC, as well as more reliable abstention behavior. Our results demonstrate that suppressing entity-linked supervision during pretraining induces a shift in epistemic behavior: KLLMs rely less on parametric knowledge and more on external evidence, leading to improved reliability under realistic conditions. This suggests that pretraining-time control over knowledge acquisition can complement retrieval-augmented and tool-based systems by providing a more evidence-sensitive base model.

Context engineering decides what information a model carries forward, and current designs meter it in tokens: compressing the past into a bounded recurrent state, keeping a key-value entry for every token, or imposing a fixed budget through a window or eviction rule. All three make the token the unit of memory even when the stream is redundant and the task depends on the distinct information it carries. Building on a companion mechanism paper that opens a cache slot only when an incoming key is novel, so memory scales with the number of distinct items rather than tokens, we develop that allocate-on-novelty cache as a working-memory component and organize context by how a task depends on the past: recall-carried information belongs in a content-addressed novelty cache, summary-carried information in a recurrent state, and locality-carried information in a recency window. The claim is empirical and bounded. On a matched character-level control, novelty-gated attention reaches full-attention performance while attending to about half the tokens, and coupling the cache with a state-space summary matches full-attention coupling at that reduced cost; the advantage grows as context lengthens, while a sliding window is preferable on short, locality-dominated spans. On next-code prediction over synthetic Medicare claims the coupled component leads full attention and every fixed-budget eviction policy at a thousand-event horizon, whereas cost forecasting over the same stream is summary-carried and the cache is neutral. The retained memory is an inspectable table of templates, codes, drugs, or places rather than an opaque state. The experiments are small-scale and use only public data; they establish the primitive that context can scale with distinct information rather than tokens, in a working memory that is content-addressable and auditable.

Large language models (LLMs) store factual knowledge in their parameters. While recent work has shown that this knowledge resides in MLP layers, existing constructive and mechanistic interpretability models of fact-storage in LLMs fail to explain the surprising empirical phenomenon that they store facts at an information-theoretically optimal rate. In this work, we develop a theoretical account of this phenomenon. We develop the first Transformer-compatible fact-storing MLP closed-form construction that satisfies the following three properties empirically observed in LLMs: it (i) attains optimal fact storage scaling, (ii) handles arbitrary input/output geometries, and (iii) works inside Transformers. Key to our work is to analyze the decoding margin of MLPs, whereas prior work only studies MLP fact storage. Under isotropic embeddings, our construction achieves information-theoretically optimal storage capacity scaling and requires $10$-$104\times$ fewer parameters at matched fact count than prior constructions. For arbitrary key and value embeddings, we show that our construction attains the same storage capacity scaling, up to penalization factors depending on the embedding geometries. Moreover, we demonstrate that our constructed MLPs can be used within Transformer blocks for factual recall tasks at optimal capacity scaling, requiring $15$-$63\times$ fewer parameters at matched fact count than prior constructions. Finally, as a proof-of-concept, we show that fact-storing MLPs enable modular fact editing by swapping a Transformer's MLP with a new one.

Retrieval-augmented generation (RAG) under a fixed reader-context budget forces a selection problem: of the evidence retrieved, only a fraction can be shown to the reader. We argue that document recall -- the standard retrieval metric -- is the wrong quantity to optimize in this regime, and we make two contributions. First, as a general contribution, we introduce answer-in-context, a diagnostic that measures whether a gold answer survives as a contiguous span in the packed reader context (not the retrieved set). It predicts answer F1 better than recall (r=0.39-0.55 vs. about 0.31), separates answer quality roughly five-fold (0.60 vs. 0.12 on HotpotQA), and carries information beyond retrieval: it adds Delta R squared=0.17 over recall and shows a 4.6x EM gap even among questions where all gold was retrieved. We also confirm it interventionally: on 2WikiMultiHopQA a packing change that raises coverage but not answer-in-context yields no accuracy gain. Second, as a conditional contribution, we cast reader-context construction as budgeted monotone submodular maximization and build a packer that jointly optimizes relevance, query coverage, representativeness, and diversity. On HotpotQA with a 160-token budget and a 3B reader it beats a strong focused heuristic, MMR, and naive packing -- by up to +5.1 F1 at equal-or-lower token cost, across three seeds. Crucially, we map the scope of this win honestly: it requires the conjunction of (i) multi-hop complementary structure, (ii) retrieval that surfaces the evidence, (iii) a binding but not extreme budget, and (iv) a reader weak enough that evidence density, not reading capacity, is the bottleneck. A quantization-controlled reader-scale ladder (3B to 7B to 14B) shows the edge over the heuristic is absorbed by 7B and significantly reverses by 14B, while the diagnostic explains every boundary with a single variable.

Conversational data is increasingly used as a persistent source of user state for long-running assistants and AI agents. However, querying this data remains challenging because conversations naturally evolve: plans are revised, preferences change, and later messages frequently supersede or contradict earlier information. Existing long-memory pipelines largely treat memories as independent text or vector objects. This approach often retrieves semantically similar but stale evidence, offering limited support for state-aware reasoning. To address this problem, we present TRACE, a query processing framework over temporal evidence graphs for evolving conversational data. TRACE models conversations as a hierarchical graph spanning events, sessions, and topics, enriched with typed temporal, causal, update, and contradiction relations. Crucially, the framework maintains validity annotations so obsolete facts remain accessible for historical queries but are discounted for current-state answers. At query time, TRACE combines vector-based note retrieval with graph-guided evidence search, generating validity-aware support paths and a hybrid context for answer generation. This design separates lexical recall from evidence reconstruction, enabling bounded query-time reasoning over long conversational histories. Experiments on long-conversation query-answering (QA) benchmarks show that TRACE improves temporal and multi-hop reasoning, with ablations highlighting the importance of hierarchy, update-aware seeding, and path-grounded evidence.

Long-running language agents need mechanisms for deciding which experiences should persist after the working context is gone. Retrieval systems can reinsert past text, but they do not by themselves show that an experience has been selectively consolidated into the model's own behavior. We introduce EVAF, an Echo-Valence Attractor Field mechanism for gated LoRA consolidation, and a test-retest protocol for measuring selective parametric consolidation under controlled interference. Across GPT-2 and TinyLlama, EVAF preferentially consolidates high-valence, high-surprise experiences while preserving retrieval-accessible factual memory through a complementary routed memory path. Test-retest measurements show stronger post-interference behavioral persistence than frozen, retrieval-only, and ungated continual-update baselines, while keeping parameter drift and cross-persona contamination low. The results support a separation between memory access and memory depth: retrieving a fact and internalizing an experience are distinct computational operations.

Automatic mapping between disease classification systems, such as the International Classification of Diseases (ICD), is a challenging yet essential task for integrating health data and conducting longitudinal data analysis. Existing embedding-based methods primarily focus on \emph{one-to-one} mappings, overlooking more complex \emph{one-to-many} scenarios. The threshold-based and top-K methods offer natural extensions; however, they involve inherent trade-offs between \emph{precision}, \emph{recall} and \emph{mapping coverage} -- the proportion of source codes with at least one mapping to a target code. To address this challenge, we introduce a novel method, which is inspired by the \emph{blocking-and-matching} pipeline commonly used in \emph{entity resolution}. In particular, we first generate a block of candidate matches (\emph{blocking}) and then employ a large language model (LLM) to identify all valid mappings within each block (\emph{matching}). Empirically, we show that the proposed method achieves higher precision with comparable recall and broader coverage across multiple ICD version pairs (ICD-9-CM$\leftrightarrow$ICD-10-CM and ICD-10-AM$\leftrightarrow$ICD-11). Our source code and dataset is available at: https://tinyurl.com/46kyn7wp.

Recurrent delta-rule models keep a fixed-size state matrix S (d_v x d_k) that compresses all past context. The state of the art (GDN-2) gates this update with element-wise matrix erase/write masks. This is powerful but has two defects. First, both gates are computed from the incoming token alone, making the model memory-blind: it decides what to erase without seeing what it has stored. Second, value-axis coupling in the erase gate blocks the WY-form triangular chunk solver that drives efficient training -- the intra-chunk system splits into d_v independent solves, collapsing throughput to serial-recurrence cost. We introduce CARVE (Content-Aware Recurrent with Value Efficiency), which fixes both and, via a single-launch "megakernel" scheduling of the same WY-form math, trains faster than the matrix-gated baseline it replaces. The key idea is architectural: restricting all gating to the key axis makes the intra-chunk coupling independent of the value index, restoring one unmodified WY-form solve. Within this constraint, CARVE conditions both gates on a content signal read once per chunk from the chunk-boundary state and folded algebraically into each gate's low-rank projection (by associativity, U(Sq)=(US)q), giving memory-aware gating at negligible extra traffic. At init the content projections are zero, so CARVE is bit-identical to the baseline; we prove the one-chunk staleness perturbs gates by only O(1/sqrt(L)), matching a measured 0.18% deviation flat up to L=128. At 1.3B parameters / 100B FineWeb-Edu tokens on H100 (three seeds), CARVE improves every axis: WikiText perplexity 15.72 vs 15.90 (hybrid 15.41 vs 15.62), +0.63 pp average common-sense accuracy, and state-of-the-art RULER and real-world recall -- while training +1.4% faster at matched depth and +19.3% at iso-quality depth, at +13% peak memory. Backed by six formal guarantees.

Personalized language-model assistants are often evaluated through a memory lens: can a model recall preferences users have explicitly stated in dialogue? More comprehensive personalization demands a harder capability -- inferring what users care about from the multimodal traces they naturally leave behind. We introduce SocialPersona, a benchmark for evaluating whether multimodal large language models (MLLMs) can recover revealed preferences from longitudinal social-media timelines and use them in dialogue. Built from longitudinal timelines of 171 everyday, non-promotional social-media users, SocialPersona contains text, images, timestamps, and 2,597 human-verified preference tags across seven interest domains, separating stable interests from recent interests. It supports two tasks: constructing structured user profiles from multimodal context and generating responses aligned with inferred profiles. Experiments with proprietary and open-weight MLLMs show that models can identify broad interest domains, yet their performance drops on fine-grained and recent interests and degrades further when inferred profiles must be used to personalize dialogue. Together with evidence that text and images provide complementary preference signals, these results indicate that robust cross-modal, long-horizon user modeling remains a key challenge, and that SocialPersona can help measure and advance progress toward assistants that infer and act on revealed preferences.

Retrieval-augmented generation (RAG) gives agents access to accumulated knowledge, but has no model of time. When a fact changes (e.g., a function is renamed or API restructured), RAG retrieves both the stale and current value with near-identical embedding similarity. The agent then either abstains or serves the superseded fact. We show this is a structural problem: on a calibrated dataset, cosine similarity distinguishes a contradicted fact from a duplicated one with AUROC 0.59 (near chance), as contradictions are often more embedding-similar to the original than rephrased duplicates. We present MemStrata, a retrieval memory maintaining temporal validity. It stores facts like RAG, preserving static recall, but when a fact's value is contradicted, a deterministic (subject, relation, object) supersession rule retires the stale value in a bi-temporal ledger - with no similarity threshold and no LLM call. Across six benchmarks run locally with a 7B model, MemStrata ties RAG on static knowledge and reaches 0.95-1.00 accuracy on evolving knowledge (where RAG reaches 0.20-0.47). The central result is the stale-fact-error rate: when required to answer, RAG serves superseded values 15-40% of the time; MemStrata drives this to ~0%, a failure class RAG cannot avoid. MemStrata achieves this at retrieval latency (~2.1s) versus ~16-18s for LLM-reranking baselines. We release the harness, datasets, and a marker-free evaluation protocol for memory under knowledge evolution.

Parametric knowledge in Large Language Models is not equally accessible across languages. As a result, standard inference techniques often struggle to surface localized facts, leading to failures in cross-lingual knowledge transfer and consistency. In this work, we investigate techniques for accessing hidden factual knowledge by exploring cross-lingual prompting strategies. We identify four inherent dimensions of cross-lingual exploration that directly govern parametric knowledge retrieval and evaluate them on multilingual factual benchmarks covering 17 typologically diverse languages. Our results demonstrate that cross-lingual exploration significantly improves knowledge transfer and factual recall, representing a more efficient compute Pareto frontier than native-language scaling. Furthermore, we observe corresponding improvements in cross-lingual consistency, exceeding what can be explained by accuracy gains alone. Overall, our work establishes multilingual prompt exploration as a highly effective inference-time strategy for unlocking latent parametric knowledge.

Enterprise text-to-SQL systems often fail before SQL is generated: the model receives the wrong schema context. Modern warehouses contain thousands of tables, abbreviated columns, informal metrics, hidden join conventions, and permission boundaries that are not captured by raw table names. We introduce Schema-First Retrieval, a retrieval layer that embeds catalog metadata rather than warehouse rows. The system indexes five typed catalog objects, tables, columns, metrics, relationships, and query history, using object-specific text templates. At query time, it combines parallel vector search, lineage expansion, cross-encoder reranking, workload memory, and deterministic access-control gates before SQL generation. On CRUSH4SQL (1,534 questions), Schema-First Retrieval reaches 96.4% table recall@20 and cross-encoder reranking adds +11.1 points at column recall@10; against an equally-templated BM25 baseline, semantic retrieval is +32.8 points at table recall@5. On SEDE (857 questions), query history raises table recall@5 from 52.1% to 92.3%. On BIRD (96 questions), schema-first context reduces SQL execution errors from 15.6% to 6.2%, a 2.5x reduction. These results show that catalog selection is a first-class retrieval problem for natural language analytics, not a prompt formatting detail.

Vision-Language-Action (VLA) models are commonly fine-tuned through passive imitation learning, where additional demonstrations are collected for tasks where the policy performs poorly. This approach incurs several downsides: it requires the robot to fail before data collection is triggered, provides little guidance about which states require supervision, and wastes demonstrator effort on redundant parts of the task where the policy already performs well. In this paper, we propose an active, continual learning paradigm for VLAs. We demonstrate that active, uncertainty-guided data collection leads to more efficient fine-tuning than when using passively-collected demonstrations. However, we also find that fine-tuning only on actively-collected recovery data leads to catastrophic forgetting. We evaluate techniques for continual learning, including replay-based data mixing and elastic weight consolidation, and identify tradeoffs between plasticity to uncertainty-guided recovery data and retention of previously learned behaviors. Overall, our work contributes an empirical study of active continual learning for autoregressive VLAs, establishing that uncertainty-guided recovery demonstrations can improve adaptation efficiency while also revealing open challenges when targeted new data is incorporated into large robot policies.

While Large Language Models (LLMs) are increasingly deployed in long interactions, existing evaluations focus predominantly on retrospective memory (RM) via explicit queries. Prospective memory (PM), the critical ability to spontaneously recall and act on latent constraints without direct prompts, remains largely unevaluated. We introduce TriggerBench, a comprehensive PM benchmark spanning five dimensions across both daily assistants and professional workflows. TriggerBench pairs scenarios with matched RM controls, contrastive positive/negative variants, and overloaded triggers, enabling fine-grained measurement of proactive recall, false-alarm rate, and attentional robustness under a single protocol. Our evaluation yields three key findings. (i) PM shows a precision-recall trade-off and attentional fragility. Though enhanced reasoning significantly improves proactive recall, models may overfit to an "always-remind" heuristic. Furthermore, PM accuracy degrades substantially under implicit constraints or triggers overloaded by concurrent user requests, indicating that robust PM remains an open challenge. (ii) PM is notably harder than RM: on identical contexts, RM near-saturates up to 100K tokens, while PM decays sharply as context length scales. (iii) PM may serve as a behavioral probe of spare reasoning capacity. Pairing PM scenarios with AIME-2025 math problems reveals that successful trajectories yield higher PM accuracy than failed ones at the same context length, showing PM tracks spare reasoning budget that token count obscures. Project page: https://github.com/KristenZHANG/TriggerBench-Official.

Reject inference methods are widely used to mitigate survival bias in credit scoring, yet their effectiveness remains poorly understood. We systematically evaluate several such methods and uncover a structural failure mode: in a natural retraining cycle, models whose accuracy improves while recall collapses create an illusion of improvement that leads practitioners to believe the system is getting better when, in fact, its rejection quality -- the ability to correctly screen out defaulters -- is deteriorating. We then propose a controlled exploration strategy that breaks the feedback loop without statistical assumptions: the lender deliberately approves a fraction of rejected applicants and observes their true outcomes. We show that accuracy and rejection quality give opposite recommendations on whether to explore: accuracy favors no exploration, while rejection quality improves with it, confirming that standard evaluation metrics are misleading under selection bias. Even minimal exploration rates (2--5%) prove sufficient in our experiments to diagnose the severity of the feedback loop at near-zero cost. Our findings are consistent across two machine learning methods and three real-world datasets, and suggest that standard evaluation protocols are inadequate for assessing models trained under survival bias.

Graph-based retrieval-augmented generation (GraphRAG) is effective for knowledge-intensive and multi-hop query tasks; however, many existing methods primarily seed entity-based graphs and rely on implicit semantic relevance propagation. This often (i) under-retrieves when user queries are abstract and semantically sparse at the entity level, and (ii) suffers from brittle multi-hop reasoning, where noisy activations can derail entity-to-entity transitions and corrupt the inferred relation chain, yielding unreliable conclusions. To this end, we propose \texttt{FlowRAG}, a semantic-aware retrieval framework that improves both semantic recall and explicit reasoning. Specifically, \texttt{FlowRAG} constructs a quad-level heterogeneous graph over passages, summaries, sentences, and entities, where summary nodes serve as a coarse semantic hub. At retrieval time, a dual-granularity activation module combines summary--query alignment with sentence-level matching to activate relevant entities under paraphrase and abstraction robustly. We then introduce a frequency-aware weighted flow module that routes relevance through entity--passage links weighted by within-passage term frequency, pruning noisy connections and extracting high-confidence reasoning paths as an explicit logic skeleton for generation. Extensive experiments show that \texttt{FlowRAG} obtains state-of-the-art performance on complex reasoning benchmarks.

Long input sequences are central to document understanding and multi-step reasoning in Large Language Models, yet the quadratic cost of attention makes inference both memory-intensive and slow. Context distillation mitigates this by compressing contextual information into model parameters, and recent work such as Doc-to-LoRA amortizes context distillation into a single forward pass that generates one LoRA adapter per document. However, producing a single monolithic adapter for all queries leads to irrelevant-query interference, limited compositional recall, and poor scalability to long-document reasoning. To address these challenges, we propose Doc-to-Atom (Doc2Atom), a compositional parametric memory framework that decomposes each document into semantically typed knowledge atoms. Each atom is compiled into an independent micro-LoRA adapter and a provenance retrieval key. At inference time, a lightweight query router selects and assembles only the relevant atoms into a query-specific adapter, which is then injected into a frozen base model. The entire system is trained end-to-end through a multi-objective distillation framework. Experiments on six diverse QA benchmarks demonstrate that Doc2Atom outperforms Doc-to-LoRA baselines while reducing the memory cost of document internalization.

Persistent memory systems promise to make LLMs more helpful by storing user beliefs over time. We show they also make models less correct by systematically amplifying sycophancy, wherein models prioritize agreement with users over accuracy. We conduct the first systematic evaluation of this effect, introducing MIST: a benchmark of synthetically generated multi-turn conversations where users express plausible misconceptions in scientific, medical, and moral reasoning domains. Testing across three state-of-the-art memory systems and five model families reveals that memory amplifies sycophantic behavior across all conditions, with up to 25x higher sycophancy rates than in-context baselines. Error analyses suggest memory extraction as the primary culprit: lossy compression into discrete snippets encodes user misconceptions while discarding corrective context. Based on these results, we propose two lightweight mitigations that substantially reduce sycophancy while matching or exceeding memory systems at factual recall.

Recent advances in multimodal recommenders excel at feature fusion but remain opaque and inefficient decision-makers, lacking explicit reasoning and self-awareness of uncertainty. We introduce ReasonRec, a reasoning-augmented multimodal agent structured around a three-stage explicit reasoning pipeline. Specifically, we propose a reasoning-aware visual instruction tuning strategy that systematically transforms diverse recommendation tasks into unified CoT prompts, enabling the VLM to explicitly articulate intermediate decision steps. Additionally, our evidence-horizon curriculum progressively enhances the reasoning complexity to better handle cold-start and long-tail user scenarios, significantly boosting model generalization. Furthermore, the uncertainty-guided delegation mechanism empowers the agent to assess its own confidence, strategically allocating computational resources to optimize both recommendation accuracy and inference efficiency. Comprehensive experiments on four standard recommendation tasks across five real-world datasets demonstrate that ReasonRec achieves over 30% relative improvement in key ranking metrics compared to state-of-the-art multimodal recommenders. Crucially, ReasonRec substantially reduces inference latency by dynamically delegating up to 35% of queries to efficient sub-models without compromising accuracy. Extensive ablation studies further confirm that each proposed reasoning and planning mechanism individually contributes substantially to ReasonRec's overall effectiveness. Collectively, our results illustrate a clear pathway towards interpretable, adaptive, and efficient multimodal recommendation through explicit reasoning and agentic design.

Understanding where LLMs store factual knowledge is critical for hallucination mitigation. We systematically quantify Late Crystallization: factual knowledge does not gradually emerge across layers but "crystallizes" abruptly at the final layers. Across five model families (Pythia, Gemma, Qwen2.5, Llama-3.1, Mistral; 0.5--14B), 26.8%--93.4% of correct answers never enter top-10 predictions at any intermediate layer, with late emergence (>80% depth) consistent across architectures. Cross-scale (Qwen2.5-14B) and cross-benchmark (MMLU: 98.2%) results confirm generality; tuned lens rules out probe artifacts. A sentiment-classification control (0.5% for Qwen vs. 85.9% factual; 2.0% for Mistral vs. 26.8%) confirms the phenomenon is specific to factual recall. Late Crystallization yields a crystallization-guided intervention principle: CAA outperforms DoLa on moderate-crystallization models (Llama, Mistral; p<0.001), with a directionally consistent reversal on high-crystallization Qwen (+25.4% vs. +15.5% MC1, p=0.069). LayerNorm ablation shows crystallization is intrinsic to the residual stream; LN scaling (x1.2) yields +11.8% MC1 with zero inference overhead. We further reveal a Computability-Memorization Spectrum: computable knowledge crystallizes earlier (layer 22.1/28) than memorized facts (28.0/28). We release MechLens supporting five model families.

Deep neural networks (DNNs) excel in computer vision tasks given large annotated datasets. In real-world applications, however, labels are often corrupted by ambiguity, human error, or dynamic environments. Over-parameterized DNNs easily memorize these noisy labels during training, degrading model accuracy and generalization. Existing data-cleaning and sample-selection strategies often rely on manually specified thresholds, prior knowledge of the noise ratio, or a single metric (either learning dynamics or geometric structure), making them unstable in complex data regimes. This paper proposes a self-adaptive data-cleaning framework that integrates local, global, and learning dynamics cues for robust noisy-label detection. Samples are mapped into a unified low-dimensional feature space through a modular feature concatenation paradigm. We provide two instantiations: a 2D metric integrating class-adaptive KNN-based local disagreement with k-means-based global centroid distance, and a 3D multi-metric that additionally incorporates a z-normalized score. Unlike conventional 1D Gaussian Mixture Models applied to a single scalar metric, our framework performs multi-metric clustering on the feature space to adaptively partition samples into clean-dominant and noise-dominant components without requiring manual thresholds or noise priors. Experiments on CIFAR-10, MNIST, and ImageNet-100 with 5% to 40% symmetric label noise show high recall across settings, including near-perfect recall (>=98%) on ImageNet-100 at 40% noise. Subsequent training yields accuracy gains across evaluated settings, especially under severe corruption on ImageNet-100. These findings suggest that multi-metric integration provides a threshold-free, practical, and low-tuning strategy for noisy label detection.

Long-horizon LLM sessions outlive their context windows, and the standard mitigations - truncation, summarization, retrieval - share a structural flaw: they treat history as flat text, discarding precisely the content that makes a session resumable: decisions and their rationales, task status, and file modification history. We present TokenMizer, an open-source transparent proxy that maintains session history as a typed knowledge graph and, at context boundaries, replaces the raw transcript with a token-budgeted serialization of session state. The schema comprises 14 node types and 7 edge types under an 8-state lifecycle in which decisions can be superseded or explicitly invalidated; bitemporal validity intervals support time-travel queries; and first-class decision-transition records preserve why each decision replaced its predecessor (trigger, reason, evidence). Version 0.3.1 embeds this memory core in a production-shaped serving layer - SSE streaming, security middleware, nine provider adapters, a monitoring dashboard, graph exports (D3 JSON, self-contained interactive HTML, Obsidian Canvas) - and exposes checkpoint/resume to agents as Model Context Protocol tools. The evaluation is deliberately minimal and fully provenanced: three synthetic sessions, heuristic-only extraction, one plain-summary baseline, every value traceable to a single versioned results file. Graph extraction ties the baseline on task recall (75.6%) and exceeds it on decision recall (85.0% vs. 70.0%) and file recall (100% vs. 91.7%), with 201-302-token resume blocks extracted in 8.1-529.9 ms per session. At n=3 these results are directional; ceiling effects and baseline weaknesses are analyzed explicitly. Code, benchmark runner, and the exact results file are released under the MIT licence.

Large language models have substantially advanced Text-to-SQL systems, yet applying them to enterprise-scale databases remains challenging. Real-world databases often contain large and heterogeneous schemas, incomplete metadata, dialect-specific SQL syntax, and complex analytical questions that are difficult to solve with a single SQL query. To address these challenges, we propose ProSPy, a Profiling-driven SQL--Python agentic framework for enterprise-scale Text-to-SQL. ProSPy structures the reasoning process into four stages: it first extracts fine-grained data evidence through automatic profiling, progressively prunes large schemas into task-relevant contexts, fetches intermediate views through a dialect-agnostic SQL interface, and finally performs flexible downstream analysis with Python. This design combines the efficiency of SQL over large databases with the flexibility of Python-based analysis, while reducing reliance on unreliable metadata and improving robustness across SQL dialects. Experiments on Spider 2.0-Lite and Spider 2.0-Snow show that ProSPy consistently outperforms strong baselines with both open-source and proprietary models, achieving execution accuracies of 60.15% and 60.51% with Claude-4.5-Opus, without majority voting. Further analysis shows that ProSPy is robust to SQL dialect variations and achieves a favorable trade-off between schema recall and precision.

Prompted knowledge cutoff instructs a large language model (LLM) to act as if information beyond a specified cutoff date were unavailable. However, prior work mainly relies on direct-answer generation, which struggles when post-cutoff knowledge is not explicitly queried but is only causally related to the question. To address this limitation, we propose two recall-based prompting strategies: Self-Recall (SR), which asks the model to restate its cutoff constraint, and Question-Recall (QR), which requires the model to recall question-relevant information valid under the cutoff. Across three existing benchmarks, our methods outperform both direct-answer prompting and conventional step-by-step reasoning baselines, with particularly strong improvements on counterfactual questions. To investigate robustness across different cutoff settings, we further construct the Multi-cutoff Historical Event Benchmark (MHEB), which evaluates the same question under multiple cutoff years. Results show that knowledge cutoff performance varies with cutoff distance, while combining SR and QR consistently yields the best performance.

Role-playing language agents (RPLAs) simulate specific characters and personas across applications such as entertainment, companionship, interactive storytelling, and education. Faithful role-play requires more than producing plausible, in-character responses: as a character's values and behavior change over a narrative, an RPLA should reflect the character's state at the relevant stage. However, existing benchmarks largely treat characters as fixed personas or test only what they know at a given point in the narrative. We introduce ArcANE (Arc-Aware Narrative Evaluation), a benchmark for evaluating whether an RPLA follows a character's development across a narrative. ArcANE first builds an Arc that maps how a character's values, motivations, or relationships change over the story. The benchmark then scores how well an RPLA's responses fit the corresponding stages of the Arc, covering three distinct scenario types: scenes from the novel, new situations within its world, and situations outside that world. We evaluate six models under six ways of providing narrative context. In every model, using the Arc up to the queried chapter yields the best performance, outperforming the strongest non-Arc context by 2.2-8.4 points. These results suggest that faithful role-play requires evolving character states and tracking their trajectory, rather than merely retrieving relevant episodic evidence.

Approximate nearest neighbor (ANN) search has become a core primitive in information retrieval and modern machine learning tasks, from classification to retrieval-augmented generation. The community evaluates and tunes ANN algorithms primarily on their throughput at a given Recall@k, the fraction of true exact neighbors retrieved. We argue that what really matters in ANN search is the quality of the retrieved results and not their overlap with the true kNN set. We show that using Recall@k to assess retrieval quality forces unnecessary computational overhead and investigate replacing it by 1/Ratio@k, the inverse approximation ratio. 1/Ratio@k evaluates the differences between the distances of the retrieved and true neighbors. It is judge-free, hyperparameter-free, and computable from standard ANN benchmark inputs alone. We benchmark state-of-the-art ANN algorithms across diverse datasets spanning a wide range of intrinsic dimensionalities, evaluating the two metrics comprehensively across efficiency, downstream classification, and retrieval-augmented generation. On the efficiency axis, optimizing for 1/Ratio@k reaches operational quality thresholds at a substantially lower computational cost than Recall@k. In downstream tasks, performance indicators (label precision, semantic similarity, BERTScore, and LLM-graded quality) remain highly stable even when Recall@k drops significantly. The inverse approximation ratio, on the other hand, closely mirrors this stability, tracking true utility much better than Recall@k. Ultimately, while Recall@k overstates the true cost of approximation, 1/Ratio@k offers a more accurate, deployable proxy for actual ANN quality.

Conversational agents that serve as lifelong companions must maintain persistent memory across all interactions. However, simply expanding context windows with raw retrieval degrades reasoning quality, while training memory agents via standard reinforcement learning creates a severe credit assignment bottleneck in a multi-stage pipeline. To solve this, we introduce SALIMORY, a framework that trains a single language model to manage a cognitively-structured memory-spanning user facts, preferences, and working memory. By introducing a hierarchical stage-wise process reward and reward-decomposed contrastive refinement, SALIMORY provides isolated supervision for distinct memory operations (selective filtering, consolidation, and cue-driven recall) end-to-end. SALIMORY cuts memory-attributed failures by one-third, outperforms the state-of-the-art by over 10% in end-to-end accuracy, and more than doubles the Good Personalization rate.

Activation patching can identify a mixture-of-experts (MoE) block whose clean output restores a corrupted factual prediction. However, because the block output combines contributions from multiple routed experts, block-level rescue does not establish whether the recovery localizes to an individual expert or depends on the routed expert set. We study this question on single-token COUNTERFACT contrasts by corrupting subject-token embeddings, restoring clean block outputs, and then restoring clean-minus-noised expert updates under fixed routing. In Qwen3-30B-A3B-Base, a discovery sweep selects layer 44, and held-out analysis identifies L44E069 as a recurrent routed contributor with positive specificity over same-layer active experts. Its effect is fact-matched and improves true-token probability and rank, which explains part of the layer rescue. In Mixtral-8x7B-v0.1, the selected recurrent singleton is not specific; matched-size controls instead show specificity of the clean-routed top-2 set. These findings show that successful MoE-block restoration does not necessarily imply localization to a single expert.

We present eMEM (Embodied Memory), a hybrid graph-based memory system for embodied agents operating in physical environments. Current agent memory architectures, such as Generative Agents, MemGPT, and A-MEM, treat memory as text streams or knowledge graphs, but embodied agents require memory that is simultaneously searchable by meaning, space, and time. eMEM fills this gap with a multi-index architecture (SQLITE for structured storage, hnswlib for approximate nearest neighbour semantic search, and an R-tree for spatial queries) unified behind a single graph model. A tiered consolidation pipeline transforms raw perceptual observations into compressed summaries, mirroring hippocampal-neocortical consolidation in biological systems. Ten agent-facing recall tools expose memory retrieval primitives, including concept-to-location resolution and cross layer recall, as first-class operations for LLM tool calling. The system is fully embedded and runs in-process alongside the agent. In addition we introduce eMEM-Bench v1, a benchmark we construct over ProcTHOR-10K scenes for embodied memory evaluation. The benchmark is organised explicitly around eight cognitive-psychology paradigms (DRM lures, pattern separation, pattern completion, source monitoring, context-dependent retrieval, long-horizon interference, serial position, and a foil augmented retention curve), each chosen so that the result is interpretable against the broader memory-systems literature in humans and prior agent-memory systems; a level of diagnostic that surface-task benchmarks like LoCoMo or OpenEQA cannot provide. eMEM scores 80.8 weighted mean over 988 probes, with a flat retention curve at ceiling from 1 h to 1 yr of simulated delay on room-unique items. We show that a pure RAG baseline (the flat_rag ablation) loses 30 pt on context dependent retrieval and 29 pt on DRM lure rejection, isolating the contribution of multi-layer storage and consolidation respectively. We release both the system and the benchmark code.