Overview

Knowledge distillation studies how a student model can learn from a teacher’s outputs, internal representations, examples, or behavior, often to reduce inference cost or transfer capabilities. Common approaches match softened class probabilities, hidden features, attention or relational structure, and use teacher-generated data; newer language-model methods also train on student-generated rollouts. The field spans model compression, cross-architecture and multimodal transfer, and settings such as federated learning, robotics, and low-resource tasks. A central challenge is choosing supervision that fits the student and remains reliable: stronger or denser teacher signals can help, but can also cause negative transfer, forgetting, or loss of capabilities not captured by the benchmark.

Latest papers

887 papers87 papers in the last four weeks, down 49% on the four weeks before. 0.7% of all new papers.

Nov 3, 2026

Agents, Think Only When it Helps

LLM agent distillation transfers agentic capabilities to student models by training on complete teacher trajectories that typically interleave thoughts and actions. We question how critical thought supervision is for agents' ultimate action performance. To answer this, we disentangle their impacts by independently perturbing teacher thoughts and actions during distillation. On widely adopted agent benchmarks across model scales, altering or discarding thoughts has little effect on student performance. We find, however, that thought supervision becomes important when information about the thought is not disclosed in the corresponding action during training. By identifying such discrepant cases, our findings challenge the common practice of distilling entire teacher trajectories for LLM agents, and offer a new insight that thought supervision can be a discernible choice for efficiency depending on whether the information of thought is disclosed in the action.
Nov 3, 2026

TRUST-Stereo: Teacher-Assisted Robust Uncertainty-Guided Search and Transfer for Fast Stereo Matching

Coarse-to-fine stereo matching enables efficient disparity estimation but remains sensitive to errors in propagated disparity, which may place ground-truth correspondences outside subsequent local search ranges. We propose TRUST-Stereo (Teacher-Assisted Robust Uncertainty-Guided Search and Transfer for Fast Stereo Matching), which addresses this limitation through an Uncertainty-Guided Residual Search Module (UGRSM). UGRSM estimates the reliability of propagated disparity and predicts residual offsets to adaptively relocate local cost-volume centers. Additionally, we introduce a two-stage Senior-Junior Teacher Assistant knowledge distillation strategy to transfer rich monocular representations from a large foundation model to a lightweight backbone to achieve fast stereo matching performance. Experiments on KITTI 2012, KITTI 2015, ETH3D, and Middlebury demonstrate strong zero-shot generalization. Using only 1.4M synthetic stereo pairs, TRUST-Stereo achieves state-of-the-art results on several benchmark metrics, outperforming competing efficient methods trained with substantially larger datasets, including real-world pseudo-labeled samples. Specifically, it achieves a KITTI 2015 EPE of 0.93, ETH3D Bad 1.0 of 2.13% and Middlebury Bad 2.0 of 6.79% while maintaining inference at approximately 50 ms per stereo pair.
Nov 3, 2026

MDSKILL: Learning to Internalize Skill Memory through Multi-View Distillation for Efficient Dynamic Agent Self-Evolution

Skill memories encode past experiences as textual prompts to improve reinforcement learning (RL) for self-evolving agents. Complementing RL, On-Policy Self-Distillation (OPSD) provides dense, fine-grained supervision to internalize knowledge from skill-guided trajectories and accelerate policy learning. However, self-generated memories often contain noisy supervision and functionally entangled, redundant, or conflicting knowledge, making direct distillation difficult. Under limited training budgets, these issues can hinder knowledge internalization and even degrade performance. To address this problem, we propose MDSKILL, a multi-view distillation approach for skill memory internalization. MDSKILL organizes skill memories into three complementary functional views (state grounding, action planning, and execution control) and selects high-quality skills using dynamic metrics to construct view-specific and integrated reasoning branches. OPSD then transfers and consolidates complementary experiential knowledge across these branches into a shared policy, enabling efficient skill internalization and continuous agent self-evolution. We evaluate MDSKILL on three agent benchmarks against RL, skill-augmented, and self-distillation baselines that do not rely on stronger external teacher models. Without additional computational overhead, MDSKILL improves upon SOTA performance by up to 7.1% and 8.1% on ALFWorld and WebShop. Extensive ablations and evolution analyses further validate its effectiveness in knowledge internalization and policy improvement. The code is available at https://anonymous.4open.science/r/MDSKILL-CF53.
Nov 3, 2026

Future Representation Distillation for Imagination-free World Action Models

World Action Models (WAMs) use the predictive knowledge of pretrained video diffusion models for robot control through future video prediction. However, iterative video denoising is costly, and inaccurate visual predictions can compromise action prediction. Our key idea is to make this predictive knowledge available through visual representations computed from the current observation, enabling action prediction without generating future videos. We introduce DreamFree, a WAM that repurposes a pretrained video diffusion model as a predictive visual encoder through representation distillation. The visual encoder learns to predict a frozen video diffusion model's representation using only the current observation. An action module is then trained to predict actions by extracting action-relevant information from the visual encoder. Across real-world and simulated tasks, DreamFree achieves higher success rates than state-of-the-art VLAs and WAMs, while performing real-time closed-loop control at 11Hz with a 14B video diffusion backbone. Experiments show that DreamFree can translate stronger video priors and additional video data into better robot control and enable learning new robot tasks from video demonstrations, including videos from a different embodiment, without action annotation for the new tasks.
Nov 3, 2026

All-For-One: Distilling Visual Foundation Models to a Fast, Memory-Efficient Visuomotor Policy

Visual Foundation Models (VFMs) provide semantic cues such as depth and segmentation useful for robotic manipulation, but directly deploying them in closed-loop control is impractical due to high computational cost, temporal inconsistency, and distribution shift between static internet images and egocentric robot observations. We propose All-For-One (AFO), a two-stage distillation framework that leverages VFMs strictly as privileged information during training, using downstream robotic data. Following the privileged learning paradigm, AFO decouples semantic representation learning from inference-time policy execution. In the All stage, a teacher policy fuses RGB observations with outputs from multiple VFMs to learn a semantically enriched representation. In the For-One stage, this knowledge is distilled in-distribution into a lightweight RGB-only student policy. Evaluated on two Adroit tasks and five MetaWorld tasks in simulation, as well as three real-world manipulation tasks, AFO outperforms prior foundation-model-based approaches while using over 30x fewer parameters than VFM-based encoders and 50x lower latency than its own teacher in simulation benchmarks, providing an efficient pathway for transferring foundation-model knowledge to robotic control.
Nov 3, 2026

Where Do Distilled Students Store Thinking?

Small students degrade on large reasoning models' (LRMs') long solutions, so recent work shortens them by interpolating an LRM with a short-answer model at ratio λλ. We ask what student weights retain when this teacher stops thinking, comparing LoRA adapters of three students from three teacher pairs at each λλ. In each pair, the adapters' update subspace shifts most where the teacher stops thinking and solutions shorten most, locating the change without attributing it. We identify a shared adapter subspace tied to thinking-like behavior (reflective markers, monologue format and long solutions). Shared by adapters before that point, it keeps about half (post hoc) while one pair's teacher halves its solutions but keeps its think tag, and largely empties after thinking stops. Removing it largely turns this behavior off for every pair, student and benchmark. Only in that pair and one other student does it leave clearly fewer markers than removing the update's top components; elsewhere, same-energy random directions often nearly match it. Grafted into a short-answer student at twice its weight, it turns early markers (a post hoc count) on, like the whole update, but the student rarely finishes. Our work identifies a weight subspace tied to distilled thinking-like behavior.
Nov 2, 2026

RAPID: Increasing Efficiency of Internal Distillation via Randomized Methods

Knowledge distillation is a technique used to transfer the expertise of large teacher models into small, efficient student models. Recent advances have integrated the distillation of internal activations, more directly training students to learn similar representations as the teacher. One of the core challenges of internal distillation is the dimensional incompatibility between the student and teacher internal representations. Existing methods add learnable matrices to project the larger teacher activations into the smaller representation space of the student. This is a notable point of inefficiency, as learning additional parameters for the projectors increases training costs, only for them to be discarded immediately after training. We introduce RAPID (RAndomized Projection for Internal Distillation) a novel distillation methodology which eliminates the wasteful projection learning by leveraging randomized methods. RAPID utilizes an efficient transform construction based on the Johnson-Lindenstrauss lemma to compress the teacher representations into the student embedding space with strong theoretical guarantees of structural preservation. RAPID also employs a combination of exponential moving average (EMA) and homoscedastic uncertainty weighting in a novel multi-objective function designed for internal distillation. Our experiments in post-training for reasoning demonstrate that RAPID increases training efficiency by as much as 21.1% while maintaining performance comparable to existing methods.
Nov 2, 2026

Contrastive Adapter Training and Consensus Knowledge Distillation for Multi-Source-Free Domain Adaptation in Skin Cancer Diagnosis

Skin cancer diagnosis, particularly differentiating melanoma from benign nevi, is challenging due to the visual similarity between lesions. Although convolutional neural networks (CNNs) and vision transformers (ViTs) have shown promising performance, their effectiveness often deteriorates on data from heterogeneous clinical sources. Conventional domain adaptation methods address domain shift but require access to source data during adaptation, which is often infeasible due to privacy regulations. Multi-source-free unsupervised domain adaptation (MSFDA) addresses this limitation by leveraging multiple labeled source domains to generalize to an unlabeled target domain without requiring access to source data. However, existing MSFDA methods rely on full backbone fine-tuning, leading to catastrophic forgetting and overfitting on small clinical datasets, and address domain shift at the aggregation stage without establishing a shared domain-invariant feature space. Furthermore, their reliance on hard pseudo-labels or confidence-weighted aggregation introduces noisy supervision signals under domain shift. To address these limitations, we propose CAT-CKD, consisting of two components: (1) contrastive adapter training (CAT), which trains lightweight ConvPass adapters within a frozen ViT backbone using supervised contrastive learning (SCL) to establish a shared domain-invariant feature space before source-specific model training, and (2) consensus knowledge distillation (CKD), which aggregates logits from multiple source models into a consensus supervisory signal and adapts a student model on unlabeled target data using KL divergence. Experiments on five publicly available skin lesion datasets show that CAT-CKD achieves an average AUROC of 86.1%, outperforming existing MSFDA methods while requiring only 4.3M trainable parameters. The code is available at https://github.com/A-Abedi/CAT_CKD.
Nov 2, 2026

Meet Students Where They Are: Student-Prefix Teacher Adaptation for On-Policy Distillation

On-policy distillation (OPD) mitigates the student's exposure bias by optimizing on student-generated trajectories. However, student-generated prefixes may contain reasoning errors or follow patterns unfamiliar to the teacher, undermining the reliability of its token-level supervision. This degradation becomes more pronounced as prefixes grow longer. To address this teacher-side distribution mismatch, we propose Student-Prefix Teacher Adaptation (SPTA), which adapts the teacher to the states on which it is queried during distillation. Specifically, before OPD starts, we collect student rollout prefixes with random lengths and continually train the teacher to continue from the prefixes using reinforcement learning with verifiable rewards (RLVR). By treating each prefix as context and rewarding correct final answers, the teacher learns to recover from student-induced states rather than imitate the student's reasoning. Then, the adapted teacher is used for standard OPD, with no changes to the distillation objective. Experiments on mathematical reasoning benchmarks show that our SPTA improves average student performance over standard OPD and OPD with task-level RLVR-adapted teachers. These results highlight the importance of adapting teachers not only to solve problems independently, but also to provide reliable guidance on the states their students actually visit.
Nov 2, 2026

SurgPace: Complexity-Aware Progressive Distillation for Surgical Multimodal Large Language Models

Multimodal large language models (MLLMs) have shown strong potential in robotic surgery for question answering, scene understanding, and grounded assistance, but their high computational cost limits real clinical deployment. Knowledge distillation provides a practical way to build compact surgical MLLMs, yet direct distillation is not well suited to surgical data because surgical samples vary greatly in difficulty, visual ambiguity, and required clinical context, making uniform teacher-student transfer unstable. In addition, the role of supervision changes during training: early stages require stronger teacher guidance for multimodal alignment, while later stages rely more on task supervision for effective optimization. To address these challenges, we propose SurgPace, a complexity-aware continuous curriculum framework for progressive distillation of surgical vision-language models. SurgPace estimates sample complexity using predictive confidence and Lorentz geometric consistency, and uses the fused score to guide a continuous easy-to-hard training process. We further introduce hierarchical representation alignment, which aligns the student with the teacher at the feature, logit, and probability levels, while gradually shifting the training focus from teacher imitation to task supervision. This design separates sample-level curriculum from objective-level scheduling, leading to more stable and effective distillation. We evaluate SurgPace on Surg-396K, a benchmark built from four public surgical datasets and covering five dialogue paradigms. SurgPace consistently outperforms strong MLLM distillation baselines, demonstrating its effectiveness for building compact, memory-efficient surgical MLLMs with faster decoding under controlled benchmarks. Code is available at https://github.com/hezhicheng2002/SurgPace.
Nov 2, 2026

Distilling Literary Sentiment from LLM Annotations into an Ordinal BERT Model

The scarcity of human-annotated literary corpora limits the development of domain-specific sentiment analysis models. This study investigates whether LLM-generated annotations can supervise smaller models for literary sentiment analysis. We fine-tune five BERT models using ordinal regression on 10,889 literary sentences annotated by Gemma 4 31B on a ten-point sentiment scale. The models are evaluated against 1,923 human-annotated sentences from Hemingway's "The Old Man and the Sea", alongside the teacher model and conventional baselines. BERT-large achieves a Kendall's ττ of 0.589, approaching the teacher's 0.604. Although a constant-neutral predictor achieves the lowest overall mean absolute error (MAE), BERT-base outperforms the teacher on non-neutral sentences, achieving an MAE of 0.726 versus 1.116. These findings demonstrate the potential of LLM-generated supervision for developing smaller literary sentiment models, while highlighting the limitations of aggregate error metrics on predominantly neutral datasets.
Nov 2, 2026

Beyond Policy Patches: Distilling RL Specialists into Vision-Language-Action Policies

Vision-language-action (VLA) policies can handle noncritical stages of long-horizon manipulation yet fail during short precision-critical phases. Lightweight reinforcement learning (RL) specialists can address these failures without directly applying RL to the full VLA, but retain auxiliary controllers and switching mechanisms at deployment. We study how to transfer their capabilities into a single flow-matching VLA. Direct imitation of specialist rollouts can reproduce action jitter and leave the VLA unprepared for states encountered during its own execution. We introduce REPAIR, a value-guided distillation framework that combines offline initialization with on-policy refinement. At each critical-phase observation, REPAIR queries the corresponding specialist for multiple action candidates and weights them using its critic to construct a supervision distribution. A shared weighted flow-matching objective trains the VLA first on observations from specialist-assisted rollouts, then on observations collected through its own execution. Specialists provide supervision without taking control during on-policy collection and are removed at deployment. Across six real-robot tasks with 30 evaluation trials per task, REPAIR achieves 100% success on all six tasks, with lower mean end-effector jerk than specialist-assisted execution on every task.
Nov 2, 2026

Replay for LLM Continual Learning with Multi-Teacher On-Policy Distillation

Multi-teacher on-policy distillation (MOPD) distills several domain experts into one student and is used in the post-training of recent large language models. It trains on all domains at once, so every expert must be available at the same time. In practice, capabilities often arrive one after another, and the experts of earlier domains may be gone when a new domain is learned. Distilling the experts in sequence then suffers from catastrophic forgetting of the earlier domains. Some recent pipelines store a snapshot after every stage and end with a multi-teacher recovery stage taught by all snapshots. The number of stored models then grows with the number of stages. Different from these pipelines, this paper formulates continual MOPD by what is kept from earlier stages and studies replay toward the previous student. Inside every stage, prompts of earlier domains are replayed with the student of the previous stage as their teacher, so neither earlier experts nor stage snapshots are kept. Two design choices of replay are explored. The replay budget of a stage is either shared among the earlier domains or fixed per earlier domain. Replay steps either distill the token distribution alone or add alignment of hidden states, region-pooled self-attention scores and value relations. A five-domain benchmark of logic, math, code, safety and instruction following is built from public experts of one 1.5B student. Experiments on its 14 benchmarks show that, without storing any snapshot, replay toward the previous student closes most of the gap between sequential distillation and joint MOPD. With a fixed replay share per earlier domain, it performs on par with recovery from all stored snapshots.
Nov 1, 2026

Conditional Calibration Contracts under Coupled Load: A Branch-Aware Audit of Switched Inference

Bursty serving changes both queue states and the population selected for each model. We study conditional calibration contracts as a specification to be validated, rather than a certified property of an existing router. CA-MHD combines load-coupled distillation, closed-loop replay and a cost-informed Schmitt router in a discrete-event simulator with a hardware service table. Across 40 cells (ten training seeds and four content-load coupling levels), we distinguish the original teacher, frozen wide heads on an adapted trunk, and the narrow student. A common teacher-trajectory audit has median absolute student-teacher deficit 0.06 percentage points, but does not certify the student-selected deployment support. Historical results use a kernel correctness gap, not the published SmoothECE; we separately prove selected-branch composition for a residual functional at a common fixed scale. The deployed wide branch has lower ordinal accuracy than the original teacher in all 40 cells. Same-checkpoint student-only serving has lower historical ordinal gap in 27/40 cells; routing also exposes violation tails up to 17.5 percent. Student p99 latency remains above the 1.5 ms target. Retrospective temporal recalibration improves the ordinal gap in 39/40 cells, but causal delayed-label availability is unverified and no incremental per-mode benefit is detected after multiplicity adjustment. An in-sample CUSUM detects 30/40 checkpoint changes at five historical false alarms, with no prospective false-alarm guarantee. These findings support a branch-aware audit and a scoped negative result, not demonstrated routing superiority. Independent budgets, selected-support validation and net utility under a satisfied SLA remain open.
Nov 1, 2026

Order-Only On-Policy Distillation

On-policy distillation (OPD) trains a student on its own generations by matching the teacher's token probabilities. These probabilities entangle two kinds of information: which tokens the teacher prefers, and how confident it is. The latter reflects the teacher's own capacity and calibration, which a smaller student may be unable to reproduce. Because the teacher temperature changes these values without changing the ordering, OPD is also sensitive to the temperature choice. We ask whether the teacher's ordering of candidate tokens is enough, and propose Order-Only On-Policy Distillation (O3^{3}PD). At each student-generated prefix, O3^{3}PD trains the student to reproduce the order of the teacher's top-KK tokens, without fitting their probability values. The objective is therefore invariant to the teacher temperature and needs no temperature tuning. Across six teacher--student pairs, O3^{3}PD outperforms OPD on the average over HMMT25, AIME24, and AIME25 in every pair, by up to 2.9 points, with gains of up to 5.6 points on AIME25. Other ranking losses yield similar gains, so the benefit comes from ordinal supervision itself. Analysis shows that O3^{3}PD departs further from the teacher's probabilities than OPD, yet reproduces its ranking more faithfully and yields a stronger student. The ranking, not the probability values, thus carries the knowledge that on-policy distillation needs to transfer.
Nov 1, 2026

Bernoulli On-Policy Distillation: Don't Ignore High-Confidence Student Tokens

On-policy distillation (OPD) trains a student language model on its own trajectories using token-level supervision from a stronger teacher. We show that standard sampled reverse-KL OPD systematically suppresses updates to high-confidence tokens, such that their weak gradient contribution does not necessarily imply weak or uninformative teacher supervision. We introduce Bernoulli On-Policy Distillation (Bernoulli OPD), which directly matches the teacher and student probabilities for that token. This yields a bounded probability space correction that preserves teacher-student disagreement while avoiding the additional confidence-dependent suppression induced by vanilla OPD. Fixed-rollout analyses show that Bernoulli OPD substantially restores optimization signal to high-confidence tokens while remaining sensitive to teacher-student disagreement. We further analyze the resulting model behavior through confidence distributions, answer calibration, policy movement, and training dynamics, showing that Bernoulli OPD's gains cannot be explained simply by longer reasoning, larger optimization steps, or more uniform imitation of the teacher. Code is available at https://github.com/duterscmy/Bernoulli-OPD.
Nov 1, 2026

OC-OPD: Observation-coordinated Supervision for On-Policy Distillation

Reverse and forward Kullback--Leibler (KL) divergences provide complementary supervision in on-policy distillation: reverse KL refines the student's active token rankings, and forward KL allocates mass to teacher-supported candidates. At student-generated prefixes, we form a four-observation matrix by treating the student and teacher as both candidate sources and evaluators. The diagonal self-observations measure uncertainty, while source-centered cross-observations recover the two KL objectives. We establish an exact identity equating the Fisher inverse-metric coupling of their logit gradients with the sum of these contrasts, yielding a common local descent direction for distinct positive distributions. Observation-coordinated on-policy distillation (OC-OPD) uses observations to set preferences, closed-form bargaining to coordinate gradients, and distributional discrepancy to scale the update. Across three heterogeneous student settings (non-thinking generation, thinking generation, and base-model initialization), OC-OPD achieves strong gains in mathematical reasoning and transfers effectively to out-of-distribution scientific question answering and code generation. Our code is publicly available at: https://github.com/RuoranXu/OC-OPD.
Oct 31, 2026

Label Smoothing in Disguise? Gated Knowledge Distillation for Low-Resource Bangla Generation

Gated knowledge distillation (KD) weights teacher signals per token or example, aiming to trust an imperfect teacher only where it seems reliable. We test two gates: EWAD (entropy-weighted agreement-aware distillation), which gates tokens by teacher confidence, and CHAD (counterfactual harm-aware distillation), which gates examples by gradient alignment with validation data. Across six Bangla-centric tasks, three student models (60M to about 2B parameters), and two teacher families, no gate significantly improves plain KD in 45 comparisons. KD also fails to beat cross-entropy (CE) fine-tuning on gold outputs in 40 single-seed comparisons or three-seed runs on four tasks; CE leads on every summarisation metric. The main benefit of KD is regularisation: over ten fixed epochs, CE students lose 0.030 to 0.044 token-F1 from their early-stopped checkpoint, versus 0.005 to 0.025 for distilled students. Teacher-free label smoothing closes most of this gap. In one comparison, a teacher 0.29 token-F1 worse yields students at most 0.009 worse. Thus, gating may change regularisation strength but has little teacher knowledge to select. Findings are limited to Bangla-centric tasks and small students. We will release code, data splits, predictions, and scores.
Oct 31, 2026

ZOPGA-KD: Zeroth Order PGA Guided Data Free Knowledge Distillation

Knowledge distillation (KD) enables the transfer of learned representations from a large teacher network to a small student network. In foundational KD, the student network uses the original training data of the teacher model to accomplish this task. However, due to privacy concerns, the original training data is often not available. To overcome this reliance on original training data, several data-free knowledge distillation methods have been developed. Some of these methods operate in the white-box setting, and a few methods generate synthetic data by training a generator network. White-box methods use the teacher model's parameters to obtain a synthetic dataset. To avoid relying on teacher parameters, some methods use computationally expensive generative approaches to obtain such a dataset. There also exist certain black-box methods that generate the synthetic data without accessing teacher parameters. These methods are also computationally intensive. We propose a data-free knowledge distillation method that utilizes just the teacher's output logits and does not require access to its architecture, parameters, or gradients. We generate synthetic data via Projected Gradient Ascent (PGA) starting from smooth random noise signals. Required gradients are obtained through zeroth-order gradient estimation rather than backpropagating through the teacher's parameters. The student is then trained on these synthetically generated samples. We evaluate the proposed method on several benchmark datasets and show that the distillation performance is comparable with approaches that have access to the teacher model parameters.
Oct 30, 2026

From Lookups to Lessons: Turning Retrieved Knowledge into Parametric Memory

Retrieval-augmented generation (RAG) helps large language models (LLMs) answer factual questions beyond their training data, but it adds storage and inference costs. Distilling previously retrieved information into parametric memories enables reuse without online retrieval. However, training the memory to mimic a retriever rewards accurate independent predictions, which do not necessarily improve the final output quality when combined with the LLM's existing predictions. To address this, we introduce ResMem, in which the LLM reads retrieved text offline to generate teacher targets. This design avoids storing massive token representations, unlike existing parametric memory approaches such as MLP Memory, which relies on k-nearest neighbor (kNN) datastores. Furthermore, by using the LLM as an offline teacher that reads privileged retrieved evidence alongside the input, we train a parametric memory to learn an additive logit residual. This residual aligns the predictions of a student sharing the same backbone with the teacher distribution, while self-distillation updates remain confined to the parametric memory. On a Wikipedia subset, ResMem reduces the storage retained for teacher construction by 94.7% compared with the evaluated kNN pipeline. Across five question answering datasets, ResMem raises mean token F1 from 32.83% for the frozen LLM to 34.48%. It also breaks 21% fewer of the LLM's correct answers than a kNN-trained memory. These findings validate our motivation of treating retrieval as a learning resource, with parametric memory shaped by how retrieved knowledge is used to inform the model's future predictions. Code and model weights are available at https://github.com/SSRand/From-Lookups-to-Lessons and https://huggingface.co/RandHan/ResMem-Mistral-7B-v0.3.
Oct 30, 2026

DecisionRank: On-Policy Distillation for Decision-Model Rerankers

Decision models target structured tasks by directly predicting choices or scores, avoiding autoregressive decoding latency. However, transferring a generative LLM's ranking knowledge to a decision model remains challenging: the teacher expresses preferences through a ranked list of document ID sequences, while the student determines the ranking using scores from the first-token position. To bridge this gap, we introduce DecisionRank, a decision model trained through item-space on-policy distillation (OPD). Item-space OPD constructs the student's rollout by sorting its document scores into a ranked sequence. A generative teacher takes the resulting ranking list as inputs and provides conditional feedback to directly supervise those scores. Combined with offline ranking supervision, DecisionRank achieves state-of-the-art performance on both general and reasoning-intensive retrieval benchmarks such as TREC DL, BEIR, and BRIGHT. Across four benchmarks, item-space OPD consistently outperforms alternative training objectives. Even at 1.7B parameters, DecisionRank surpasses its 27B generative teacher on TREC DL19--23. At 8B, it outperforms its generative listwise counterpart while running 24.5×24.5\times faster on TREC DL and 16.7×16.7\times faster on BRIGHT.
Oct 30, 2026

Stepwise-Querying: Distilling Reasoning without Access to Teacher Chain-of-Thought

Reasoning distillation is widely used to transfer the reasoning capabilities of large language models to smaller students through teacher-generated supervision. In the black-box setting, however, teachers may expose only final answers and brief solutions, leaving intermediate calculations and validity conditions unavailable and thus limiting the reasoning supervision available for distillation. We propose Stepwise Querying, which elicits step-specific explanations without requiring access to the teacher's internal chain-of-thought (CoT) traces. Independent queries to the same teacher request intermediate derivations, validity checks, and corrections for individual steps, the assembled solutions are then used for standard supervised fine-tuning. Across two teachers, two students, and four mathematics and code benchmarks, our method achieves the highest score in 14 of 16 comparisons with three baselines. We further introduce Stepwise Explanation Information Gain (SEIG), a diagnostic based on the log-ratio of a frozen student's probabilities of correct step completion under stepwise and global explanations. Analyses on GSM8K show that incorrect-answer trajectories can remain useful when they preserve valid intermediate reasoning, while further decomposition yields limited additional gains.
Oct 30, 2026

Distilling Knowledge from Search: Decoder Model Factory for Quantum LDPC Codes

We introduce GraphMP, a learning-based message-passing decoder for quantum low-density parity-check codes that transfers without fine-tuning across code families, code sizes, physical error rates, and noise models. Its local message passing keeps computation scalable as the decoding graph grows. We further embed GraphMP in a beam-search algorithm that progressively refines candidate error patterns, trading additional decoding time for lower logical error rates. To turn this decoding effort into lasting improvements, we propose the decoder model factory, a paradigm for training successive models on data collected by earlier models during decoding. In our implementation, verified search trajectories provide reusable training data for distilling knowledge into subsequent models. Across generations, the model factory yields a search-enhanced decoder that is broadly competitive with Relay-BP. On the largest code block, which encodes 46124612 logical qubits, it achieves a lower logical error rate and about 2020 times the decoding throughput of the standard Relay-BP baseline. These results motivate a coupled quantum-classical architecture with two interacting cycles: rapid decoding protects quantum information, while slower learning adapts the models to changing noise conditions. The long-term vision is a scalable decoder model factory that continually produces high-performance, low-latency decoders for evolving quantum hardware.
Oct 30, 2026

Task-Performance Routing for Multi-Teacher Distillation in CT Medical Image Segmentation

Pre-trained models offer complementary strengths in CT organ segmentation, but their relative accuracy varies across organs and spatial regions. Multi-teacher knowledge distillation can combine these strengths in a single student, yet global or instance-level teacher weights overlook variation within a 3D volume. We propose Task-Performance Routing (TPR), which uses region-wise segmentation errors to determine teacher weights during training without learned routing parameters. In hard regions, TPR favors teachers with lower prediction errors; in easy regions, it uses uniform weights to combine their knowledge. A hierarchical spatial partition provides fine-grained routing in shallow encoder stages and coarse-grained routing in deep stages. Experiments on BTCV, MSD-Liver, and MSD-Pancreas with two teacher pairs and two backbone architectures show higher Dice scores than the individual teachers, adaptive ensembles, and evaluated multi-teacher distillation baselines. On MSD-Pancreas, TPR achieves Dice scores of 65.40% with SwinUNETR and 68.25% with U-Net, compared with 64.38% and 67.45% for the corresponding adaptive ensembles. Inference uses the student alone, with no routing or teacher overhead. Code is available at https://github.com/EdvinCecilia/TPR_MultiTeacher.
Oct 29, 2026

Soil Moisture Estimation in Precision Agriculture: A Knowledge Distillation Approach

Accurate soil moisture estimation is critical for precision agriculture, influencing irrigation, fertilization, and harvesting strategies. Traditional methods primarily rely on meteorological and soil property data, while recent approaches incorporate visual modalities such as aerial and geospatial imagery. However, these high-resolution images are expensive and computationally demanding to analyze. In this work, we propose Soil Moisture Distillation (SMD), a Knowledge Distillation (KD)-based multimodal learning framework that leverages unimodal teachers to enhance multimodal soil moisture estimation. SMD surpasses not only traditional multimodal fusion techniques, including MIS-ME, but also outperforms unimodal models, demonstrating the benefits of guided unimodal knowledge transfer. Additionally, we conduct a comparative analysis of soil patches and crop images, showing that crop images provide superior visual cues, leading to improved prediction accuracy. SMD achieves a 2% improvement in MAPE for soil patches and 1.2% for crops over state-of-the-art baselines, highlighting the effectiveness of leveraging both meteorological and visual data through knowledge distillation for enhanced soil moisture forecasting.
Oct 29, 2026

Distilling Large-Scale Machine Learning Interatomic Potentials to Predict Long-Timescale Electrolyte Transport Properties

Accurately predicting the transport properties of liquid electrolytes is critical for next-generation battery design and requires force fields that are both quantum-mechanically accurate and fast enough for long-timescale simulation. Recent large-scale pretrained machine learning interatomic potentials (MLIPs) approach density-functional-theory accuracy, but remain too slow to access the necessary timescales to converge transport properties. We provide a knowledge distillation procedure to transfer the accuracy of a large-scale "teacher" MLIP into fast, system-specific "students", without any additional ab initio data. Our distilled models run ~40x faster, reproduce teacher-predicted structural observables, and produce stable 20 ns trajectories in days, rather than the months the teacher would require. Across a diverse set of sodium- and lithium-ion electrolytes, student-predicted self-diffusivities, ionic conductivities, and shear viscosities track experiment and outperform classical force-field predictions, establishing distillation as a practical route to accurate, long-timescale simulations necessary for high-throughput electrolyte screening.
Oct 29, 2026

YOLOF-MLD: A Multi-Level Distillation Framework for Efficient Object Detection

Efficient object detection on resource-constrained edge devices is essential for safety-critical applications such as fire-smoke monitoring. However, current lightweight frameworks, such as YOLO26, still rely on the parameter-dense multi-branch Feature Pyramid Network (FPN), preventing full exploitation of their lightweight advantage. Replacing FPN with YOLOF's single-input-single-output (SiSo) neck is a promising solution but causes substantial accuracy loss. To alleviate this, we propose YOLOF-MLD, a CPU-efficient multi-level distillation (MLD) framework that combines a YOLO26n backbone, a SiSo neck, and MLD. It aligns features at the backbone via sliced-semantics matching, neck with receptive-field matching, and response levels. On D-Fire and Annotated Fire-Smoke (AFS) datasets, YOLOF-MLD reduces the parameter count of YOLO26n by 35.7% and achieves a 62.3% speedup on Raspberry Pi 5, attaining 71.6% and 59.9% mAP@50, respectively, compared with 75.5% and 60.0% for YOLO26n. These results demonstrate a competitive accuracy--latency trade-off for edge deployment. Our code is available at https://github.com/XiaolongWU111/YOLOF---MLD/blob/main/README.md.
Oct 28, 2026

Distilling the Shape of Uncertainty: What On-Policy Distillation Learns and How to Predict It

On-policy distillation (OPD) is increasingly adopted in the post-training of large language models. However, a stronger teacher may fail to improve the student. Prior work attributes this to a lack of new knowledge, but what that knowledge is remains unclear. We show that OPD learns not only how to branch, i.e., which candidate tokens to consider, but also where to branch, i.e., where to keep reasoning paths open. The existing token-overlap metric captures only the former. We introduce the entropy shape, the sequence of next-token entropies, to characterize the latter, and propose the Entropy Shape Correlation (ESC) to measure student-teacher agreement in entropy shape. Tracking ESC throughout training and intervening on a teacher's entropy while preserving its token preferences, we empirically show that where to branch causally drives OPD gains, independently of how to branch. Finally, we formulate OPD outcome prediction as a teacher selection task and establish an evaluation protocol for it, under which pre-OPD ESC selects the best teacher for both students and outperforms existing predictors. Code is available at https://github.com/maybenotime/opd-esc.
Oct 28, 2026

Information-Theoretic Reliability--Safety Bounds and Finite-Alignment Dynamics in AI Knowledge Distillation

The reliability and safety of artificial intelligence (AI), together with their transfer and evolution through knowledge distillation, have attracted growing attention. We establish an information-theoretic bound on the reliability--safety trade-off and characterize the dynamics of this trade-off under knowledge distillation. We first prove a universal trade-off inequality that is independent of the specific model form. Its boundary is determined solely by a single parameter Kh, which quantifies the hidden layer's ability to distinguish benign from hazardous inputs. We then develop a canonical Boltzmann-machine model of AI response behavior and derive two exact relations between reliability and safety. The trade-off curve is determined by the hazard discrimination capacity K, while the overall answer tendency controls the operating point along the curve. These relations agree well with existing experimental data. In knowledge distillation, we find that teacher responses induce a renormalization of the student model's trade-off parameters. We further use a renormalization-group approach to study multigenerational distillation, deriving exact recursion relations and the associated renormalization-group-like flows for two representative direct intergenerational distillation protocols at arbitrary teacher--student alignment strength. These flows reveal fixed-point structures, long-term dynamics, and critical behavior, yielding testable predictions for the critical transition between capability loss and stable transmission.
Oct 28, 2026

What Can Firms Do in the Face of Model Distillation?

Model distillation has become a major concern in the rapidly growing generative AI discourse. In its simplest form, distillation is the phenomenon where smaller firms can use data distilled from transcripts of conversations with powerful proprietary closed-weight models to train their own models, thus achieving reasonably high model quality but at much lower costs. Besides the legitimate safety and security concerns associated with distillation, another major issue is how it can also distort marketplace outcomes. This motivates us to initiate an exploration of the problem from an economic theory lens. We consider a setting with two players: an incumbent and a distilling competitor and characterize how the incumbent's model characteristics drive the incentives around how much the competitor should distill. We also consider the impact of distillation on key market outcomes like user-side surplus and the incumbent's utility. Our key contribution, however, is to study interventions for preventing or limiting distillation attacks. In this context, we study two settings: i) when it is possible, with a certain probability (which grows in the extent of distillation conducted), to detect an attack on the incumbent's model; and ii) when a distillation attack cannot be reliably detected. In the first setting, we show that penalty-based regulations can be very effective in completely deterring distillation, although it can have potentially adverse effects on market competition and user surplus. In the latter setting, we show that the incumbent can use self-distillation as a mitigation strategy. Importantly, the effectiveness of self-distillation depends on how cost-efficiently the incumbent can deploy a new self-distilled model on the market. On the positive side, self-distillation can preserve market competition thus protecting user surplus to some extent.