Knowledge Distillation

Vision Transformers demonstrate remarkable global modeling capacity but often underperform in data-scarce regimes. Distilling convolutional inductive biases from a CNN teacher provides an effective remedy while leaving the deployed model unchanged. However, general-purpose feature distillation transfers little in this setting. In CNN-to-CNN distillation, pooling, flattening, and logit-space projections remove the spatial grid that encodes locality and translation equivariance. Unlike a convolutional student, a ViT cannot readily reconstruct this structure on its own. In this paper, we propose iBKD, a distillation framework that preserves the spatial grid throughout the entire transfer process. Its core module, the Inductive Bias Attention Module, aggregates features from all student layers onto the teacher's grid using learned weights. It then enhances structural cues through channel and deformable spatial attention and injects them via convolutional cross-attention operating directly between spatial grids rather than token sets. The module is used only during training, leaving the deployed model as an unmodified ViT with no inference overhead. Across seven Transformer backbones and six data-scarce benchmarks, iBKD consistently outperforms both locality-guidance methods and general knowledge distillation baselines, with its advantage increasing as the amount of training data decreases.

Quantization-Aware Training (QAT) enables the deployment of quantized models with minimal accuracy degradation. However, in practical scenarios, training labels are often unavailable due to privacy, copyright, or cost constraints. Knowledge Distillation (KD) is a common approach to address this challenge, but we observe that prior work combining QAT with KD suffers from a fundamental limitation: during distillation, the range mismatch between the teacher and the quantized student model induces an unattainable residual, resulting in an irreducible lower bound on the distillation loss. Motivated by this observation, we propose SQuaT (Student-Aware Quantized Teacher Features), a label-free QAT framework with KD that theoretically eliminates this lower bound by applying the student's quantization parameters to quantize the teacher's features during distillation. Through comprehensive experiments across diverse settings, we demonstrate that SQuaT consistently outperforms strong baselines, with particularly pronounced gains in extreme low-bit (e.g., 1- and 2-bit) settings. Furthermore, extensive evaluations across various model design choices show that our approach does not rely on specific architectural assumptions, making it broadly applicable across diverse architectures and quantization settings. The source code is available at https://github.com/lcdbsa522/SQuaT.

LLM agents execute heterogeneous sequences of model calls within a single task: some invocations require careful reasoning, while others are structured steps such as formatting or tool-argument construction. Prior routing methods exploit this asymmetry by assigning easy invocations to a cheaper small model and difficult ones to a large model. Such policies reduce inference cost, but they leave the small model's capability unchanged, so attainable savings remain bounded by the work the student can already solve. MERA instead improves the small model itself, using a single model invocation as the unit of adaptation. In each cycle, MERA replays failed student invocations to obtain execution-verified teacher demonstrations, distills recurring procedures into an iteratively updated SkillBook, and fine-tunes a student LoRA adapter via supervised learning and optional GRPO. Routing serves as supporting machinery for deployment: the improved student is served behind a cost-calibrated router with verifier-backed fallback, and a candidate SkillBook, adapter, or router is admitted only when joint replay preserves task quality. Empirically, four-cycle adaptation raises Qwen2.5-Coder-1.5B from 28.7% to 49.7% pass on held-out HumanEval+MBPP. Under verifier-backed fallback, the deployed policy retains 88.3% pass at 60.8% of always-Luna cost. On TAU-2, a fine-tuned Qwen3.5-2B improves from 14/35 to 18/35 and matches an unadapted 4B model. These results indicate that verifier-backed multi-cycle adaptation can increase small-model capability, rather than only routing around a fixed student.

Pre-trained black-box predictive functions encode knowledge distilled from massive datasets and extensive computation. However, when the available input features differ from those the black box expects, direct use is infeasible. We introduce a method for transferring predictive knowledge from the black box to a new, heterogeneous input space. Our approach decomposes the target regression function into a transferable component, which the black box can inform, and a non-transferable component, which captures information unique to the new space. We propose a two-step neural network procedure, estimating the transferable component from abundant unlabeled feature pairs that bridge the two input spaces and the non-transferable component from limited labels. We derive prediction risk bounds that improve on those of a non-transfer alternative when the non-transferable component is small or smooth, and the procedure adapts to either case. Under additional conditions, the worst-case risk of our estimator is of strictly smaller polynomial order than the minimax risk of estimation from the labeled data alone. We extend the framework to multiple black boxes, each on its own input space, and show that aggregation can reduce prediction error relative to the best single black box. Simulated and real data demonstrate the practical value of the method.

On-policy distillation (OPD) has emerged as a core component of modern LLM post-training pipelines, yet we reveal a failure mode: degenerate agreement, where students exploit repetitive loops to achieve near-perfect token agreement with the teacher despite globally flawed responses. We therefore shift our focus from agreement to teacher-student mismatch, and find that mismatch tokens can be mainly categorized into two types: student-excess tokens and student-deficit tokens. Student-excess tokens are generated by the student but assigned near-zero probability by the teacher; their log-ratio corrections grow unbounded and destabilize the update. Student-deficit tokens, in contrast, are preferred by the teacher but rarely sampled by the student; their absence blocks the transfer of the teacher's reasoning patterns. To tackle these mismatch directions, we propose TIDE (Token-level Independent Deficit-Excess correction), which applies bounded Hellinger shaping to suppress the most severe sampled excesses and an analytic teacher top-$K$ injection to restore deficient probability mass without requiring deficit tokens to be sampled. Across mathematical reasoning benchmarks with multiple Qwen3 teacher-student pairs, TIDE consistently outperforms standard OPD and recent token-selection and reward-shaping baselines. Moreover, the gains of TIDE are more pronounced under strong teacher-student mismatch, where it improves Avg@8 from 6.9% to 20.3%, reduces average response length by a factor of 3.6, and substantially reduces formatting failures. Code is available at https://github.com/yzc-666/TIDE

Industrial anomaly detection (IAD) requires identifying fine-grained deviations from normal visual patterns. Multimodal large language models (MLLMs) can improve recognition accuracy by comparing query images with references at inference time, but these benefits rely on additional retrieval and processing. We investigate whether the benefits of reference comparison can instead be internalized in the model parameters. Access to references during training allows a reference-aware teacher to supervise a query-only student. However, the teacher may favor plausible responses based on query cues or language priors rather than valid visual information. We propose ADOPD, a reference-privileged on-policy distillation framework. The teacher evaluates student-generated rollouts under matched and mismatched references. The matched-reference teacher-to-student log-ratio defines the token-level learning direction, specifying what the student should learn. The likelihood gap between the two reference views estimates reference-specific support and calibrates the sequence-level weight. ADOPD achieves 77.31% average accuracy on the MMAD benchmark under zero-shot inference, improving the Qwen3-VL-4B backbone by 6.14 points and outperforming its one-shot setting by 2.64 points. Experiments show that ADOPD learns a fine-grained anomaly inspection strategy from reference comparison. The project will be available at https://github.com/withTai/ADOPD.

On-policy self-distillation (OPSD) converts feedback into dense token-level supervision on student-generated trajectories, complementing reinforcement learning with sparse outcome rewards. Its self-teacher, derived from the student's current or exponentially averaged parameters and conditioned on additional context, evolves alongside the student and its rollout context distribution. The benefit of modifying this moving target depends on how target--student probability mismatches translate into updates. We propose Self-Referenced On-Policy Self-Distillation (SR-OPSD), which constructs a normalized geometric target from the self-teacher and a frozen initial policy, then minimizes the forward Rényi divergence from this target to the student. The interpolation coefficient controls the self-teacher's contribution, while the Rényi order controls the power weighting of target-to-student probability ratios in the gradient. For fixed contexts and target components, we establish a conditional variational characterization and derive the exact token-logit gradient, revealing how anchoring and projection jointly shape the effective update target. Experiments across scientific reasoning, tool use, mathematical reasoning, and code generation demonstrate strong performance across multiple model families and scales. Ablations further show that reference anchoring can improve or degrade performance depending on the projection objective, supporting the joint design of target construction and projection geometry.

External natural-language skills provide large language model (LLM) agents with reusable and editable guidance for solving complex tasks. Yet their effectiveness depends not only on skill quality, but also on whether the policy can translate the provided guidance into appropriate actions. However, methods specifically designed to improve this skill-utilization ability remain largely underexplored. In practice, skill-based agents are commonly trained with reinforcement learning objectives centered on task-level rewards, which offer limited supervision and struggle to capture subtle differences in how effectively the policy uses the provided skills. We propose BCSD (Bidirectional Context Self-Distillation), a framework that combines self-distillation with reinforcement learning to train LLM agents to use external skills more effectively. Unlike prior self-distillation methods that rely on a single privileged context, BCSD evaluates each trajectory from two complementary skill-context views. The augmented view introduces higher-level Meta-Skill guidance, while the reduced view prunes general guidance to highlight task-specific skills. Their complementary token-level signals are combined to rescale the RL advantage. Experiments on ALFWorld and WebShop demonstrate that BCSD achieves the strongest overall performance across model scales, enabling agents to utilize external skills more effectively. Ablation studies further verify the complementary contributions of the augmented and reduced context views. Code will be released to ensure full reproducibility.

Data-Free Knowledge Distillation (DFKD) transfers knowledge from a pretrained teacher model to a compact student model by synthesizing semantically informative data, eliminating the need for access to the original training dataset. Existing DFKD methods rely heavily on architecture-specific statistical priors (e.g., Batch Normalization statistics) to guide data synthesis, however, such architecture-dependent priors are often absent in modern architectures such as Vision Transformers (ViTs), resulting in degraded semantic quality of the synthesized data and consequently catastrophic performance degradation. In this paper, we propose UniDFKD, a unified data-free knowledge distillation framework that replaces architecture-specific statistics with explicit, architecture-agnostic semantic priors. UniDFKD governs the entire synthesis-distillation pipeline along three dimensions: (1) Categorical Semantic Conditioning (CSC) defines what to synthesize by persistently modulating the generator with language-derived embeddings to capture semantic diversity; (2) Spatial Semantic Anchoring (SSA) dictates where evidence belongs by anchoring the teacher's spatial attributions to a Gaussian prior; and (3) Spatial Semantic Distillation (SSD) controls how knowledge is transferred by explicitly aligning teacher-student spatial evidence alongside predictions. Extensive experiments across CNNs and ViTs demonstrate that UniDFKD establishes a new state-of-the-art, outperforming existing methods by an average absolute margin of over 20% in both homogeneous and heterogeneous settings.

On-policy self-distillation (OPSD) uses a privileged teacher to supervise a reasoning model on prefixes sampled from its own rollouts. Yet each rollout also reveals how the student's response unfolds and whether it succeeds, student-specific hindsight that standard OPSD does not use to form the teacher. We introduce Privileged Adaptation from Student Trajectories (PAST), which treats each completed student trajectory as additional privileged information for the OPSD teacher while leaving the student's distillation prefixes unchanged. PAST preserves the student's next-token distribution on correct trajectories and uses failed trajectories to adapt the teacher toward verified success under student-proximity regularization. We characterize what such a trajectory-conditioned teacher can transfer to a prefix-only student. Forward-KL distillation projects the teacher distributions to their conditional arithmetic mean given the prefix. This projection separates trajectory-specific variation that remains privileged from the mean policy shift available to the student. For correct trajectories, the unclipped population objective also has the frozen student as an ideal distributional fixed point. Across three mathematical reasoning benchmarks, PAST improves the Avg@12 macro average over Vanilla OPSD by 5.6 percentage points. A $2\times2$ factorial study shows gains from both complete-trajectory access and teacher adaptation, while trajectory removal and shuffling confirm that the adapted teacher uses the matching hindsight context.

Federated Learning is a distributed machine learning paradigm that trains a global model by aggregating local clients without sharing private data of each client. Federated Distillation (FD) builds upon this paradigm by leveraging knowledge distillation to exchange soft predictions on proxy data instead of model parameters, enabling more efficient communication and supporting heterogeneous model collaboration. However, FD models trained on In-Distribution data are hardly adapted to Out-of-Distribution (OOD) scenarios. In this paper, we propose a domain-aware proxy selection framework to better adopt proxy data for OOD problems. The experimental results show that the proposed models effectively address the challenges of distribution shifts under OOD with and without proxy data by achieving average 82.9% and 80.6% over existing works on standard benchmarks. The codes and data are released in https://anonymous.4open.science/r/DPS-FD-8596.

Knowledge distillation trains a smaller student to match the outputs of a larger teacher. Feature-based methods also align intermediate representations, but this extra constraint may affect students differently. We study this question on CIFAR-100 using a ResNet-50 teacher, a width-controlled CustomResNet family and MobileNetV2 as a cross-design comparison. For each student, we evaluate each feature method against a matched logit-KD run using the same teacher, optimizer settings, training schedule and seed. We repeat the main comparisons across multiple seeds. Logit KD improved every tested student over its scratch baseline. Attention Transfer showed no clear relationship with size inside the CustomResNet family, but its average effect was negative for that family and positive for MobileNetV2. FitNets was below logit KD in all 15 paired runs. Within the constant-depth width sweep, its gap increased for wider students, although the different-depth w=48 student did not follow this trend. Finally, the same auxiliary coefficient produced different gradient scales across students, showing that a fixed coefficient does not create a uniform training condition.

On-policy self-distillation (OPSD) improves the reasoning abilities of LLMs by internalizing privileged context into model parameters through self-distillation. Two recent research lines promote vanilla OPSD by choosing which tokens to learn from and by controlling how much privileged information the teacher receives, respectively. However, we show that each line optimizes one variable while holding the other fixed, which leads to a suboptimal solution. We argue that the two variables are coupled through the student's learning capacity: the privileged information sets the per-token divergence the teacher prescribes, while token weighting selects which of these the student must absorb. We formalize the two lines of work into a unified optimization framework, which maximizes the aggregate teacher--student divergence, subject to a budget on the aggregate learning difficulty the student can absorb. Under this modelling, we propose Unified On-Policy Self-Distillation (USD), a lightweight online algorithm to solve the Lagrangian. USD reveals that a single dual variable governs both decisions: at one price for learning difficulty, it simultaneously sets the token-selection threshold and the direction of privileged-information adjustment, keeping supervision matched to the student's evolving capacity. Through extensive experiments, USD consistently demonstrates superior performance over OPSD and token- and PI-side baselines across various model scales on various reasoning benchmarks. Code is available at https://github.com/lauvlalala/USD.

Foundation segmentation models can provide supervision for spacecraft imagery without manual training masks, but their predictions vary with textual prompts and may contain geometric errors that are amplified during distillation. This paper presents GeoDistill-Refine, a two-stage framework that transfers offline SAM 3 pseudo-masks to a compact segmentation network. Six fixed prompts are fused by an unweighted 50% vote to stabilize the teacher output. The student first learns the foreground silhouette and is then refined with signed-distance-field, skeleton, and area objectives derived from the pseudo-mask. A sample-level gate, computed from prompt agreement, the valid-prompt ratio, and pseudo-mask area plausibility, reduces the influence of unreliable pseudo-geometry. On the SpaceSense-Bench HJM lockbox set, GeoDistill-Refine improves Image IoU and Boundary F1 by 0.0456 and 0.1380, respectively, over a plain pseudo-label student. External evaluations on the SPEED+ Lightbox and Sunlamp domains and on TANGO show competitive regional overlap together with gains in boundary quality or foreground precision. The deployed TinyUNet contains 0.263 M parameters and requires approximately 1.1 ms per image on an RTX 4090; SAM 3 pseudo-mask construction and the auxiliary geometry branches are used only during training.

Memory systems have shown promise for improving agent performance, but their potential remains largely unexplored for small language models, which struggle to generate sufficient successful trajectories on their own. We propose Agent Memory Distillation (AMD), a training-free framework that transfers structured knowledge from a large teacher agent to a small student agent through hierarchical memory. AMD constructs three complementary memory types from successful teacher trajectories: Workflow memory encodes task-level strategies, Subtask memory provides concrete behavioral examples at an intermediate granularity, and Function memory captures per-function calling conventions and common pitfalls. Workflow and Subtask memories are injected proactively at the start of each task, while Function memory is retrieved reactively upon tool-calling errors. We evaluate AMD on three tool-use benchmarks using four student models (4B-8B parameters) with GPT-5-mini as the teacher, achieving average accuracy gains of 27.2%p, 11.2%p, and 3.4%p on AppWorld, BFCL V3, and ToolSandbox, while consistently outperforming existing memory-based baselines. Further analysis shows that Subtask memory contributes the largest gains, teacher effectiveness depends on both teacher capability and student compatibility, and 4B-sized students benefit most from AMD.

We present InsertFuse, a unified framework for multi-category reference-guided image insertion. Its key idea is to decouple category-specific expertise learning from cross-category capability consolidation. InsertFuse first trains specialized experts for different insertion categories and then introduces Insertion On-Policy Distillation (IOPD) to consolidate their capabilities into a single student. By querying the matched expert at states visited by the student, IOPD preserves category-specific insertion behavior while mitigating the cross-category interference caused by direct joint training. To improve spatial control, we propose Token-Aligned Geometry Conditioning (TAGC), which maps mask-derived geometric cues to the visual token grid, and Region-Balanced Flow Matching, which separately normalizes prediction errors inside and outside the insertion region to prevent background-dominated and scale-dependent supervision. We further introduce Reference CFG to isolate and strengthen the guidance induced by the visual reference under fixed scene and geometry conditions, with IOPD transferring this enhanced supervision into the unified student. Extensive experiments on the public AnyInsertion benchmark and our multi-category test set demonstrate state-of-the-art performance on most metrics, showing strong reference fidelity and generation quality across diverse insertion categories.

In modern distributed network environments, particularly in Internet of Things infrastructures and 5G networks, stringent privacy preservation and scalability requirements have created significant challenges for intrusion detection systems. Although federated learning preserves privacy by preventing data centralization, its efficiency and stability is considerably degraded under severe statistical heterogeneity and resource constraints of edge nodes. To address these limitations, this study introduces the FedTransKD-IDS framework, which enhances both system stability and efficiency by integrating robust aggregation based on the geometric mean, federated transfer learning, and knowledge distillation. Within this framework, the collaboratively trained global teacher model transfers its feature extraction component to lightweight student models. Experimental evaluation on heterogeneous datasets demonstrates a peak detection performance, achieving an accuracy of 99. 18% and a recall of 99. 99%, thereby indicating the effectiveness of structured knowledge transfer in federated environments.

GUI agents are commonly trained offline from successful interaction trajectories. Standard training decomposes each trajectory into prefix-action pairs: the agent predicts an action from the current screen and interaction history, while the subsequent observation is discarded. This removes the rationale of why an action is correct: the evidence often appears only on the subsequent screen. For example, to enable Soft Wrap, the agent should click Edit or View, but nothing reveals this until the menu opens. Without such evidence, standard imitation gives the model little chance of ever sampling and thus learning the correct reasoning. To address this issue, we propose Gated Hindsight Distillation (GHD), which uses the next screenshot as privileged information during training. A student predicts from the observable trajectory prefix, while a parameter-sharing teacher additionally observes the next screenshot and re-scores the student's on-policy responses. We apply distillation only when the student fails and the hindsight-conditioned teacher recovers the demonstrated action. GHD improves task success over GRPO on AndroidWorld and AndroidLab across two vision-language models. The code and checkpoints will be made available.

To address the limitations of video-based emotion recognition under ambiguous or socially masked behavioral cues, as well as the poor deployability of physiological signals, this paper proposes a reliability-aware physiology-to-video knowledge distillation framework, termed BioKD. The proposed framework leverages physiological signals as privileged information during training to guide a video-based student model in learning deep affective representations, while relying solely on non-intrusive video inputs at inference time. To cope with the high noise and instability of physiological teacher supervision caused by inter-subject variability, signal artifacts, and temporal inconsistency, BioKD incorporates a sample-wise reliability-aware gating mechanism together with a progressive distillation strategy. By adaptively regulating the strength of knowledge transfer, the framework suppresses negative transfer induced by unreliable physiological supervision and enables more stable cross-modal distillation. Experiments on DEAP and AMIGOS show that BioKD consistently outperforms representative baselines under both trial-wise and subject-wise evaluation protocols for valence and arousal recognition. For example, BioKD achieves 68.01% on DEAP (trial-wise arousal) and 65.29% under the more challenging subject-wise setting, demonstrating improved performance under a subject-independent evaluation setting. Further analyses show that BioKD effectively mitigates overconfident teacher errors and outperforms an entropy-only weighting strategy, confirming the importance of explicitly modeling supervision reliability. In addition, BioKD introduces no additional inference-time overhead relative to the same video student architecture and removes the need for physiological sensing and multimodal synchronization.

Knowledge distillation for image restoration typically aligns intermediate features or relation matrices between teacher and student networks as static targets, ignoring the dynamic structure of the knowledge transfer process. In this paper, we propose Flow-Map Distillation on Relation Manifolds (FoRM), which reformulates relation-based knowledge transfer as a continuous flow mapping problem on the relation manifold. Rather than regressing a constant velocity field between student and teacher relation states, FoRM learns a flow map operator $\mathcal{F}_θ(\mathbf{z}, t, s)$ that directly predicts the relation state at any target time $s$ given the current state at time $t$, enabling richer trajectory-level supervision. To ensure global self-consistency of the learned flow map, we introduce a safe semigroup consistency constraint that enforces compositional agreement using ground-truth bridge states, eliminating phantom-state error accumulation. An endpoint anchoring loss further prevents the operator from drifting away from the teacher target. Extensive experiments on five image restoration tasks, including super-resolution, deraining, denoising, deblurring, and low-light enhancement, demonstrate consistent gains over state-of-the-art distillation baselines across multiple backbone architectures, reducing training variance by approximately 50% compared to naive flow matching distillation while achieving superior restoration quality.

Subliminal Learning (SL) is a surprising type of generalization displayed by modern language models. It allows the transfer of a bias or behavior from a teacher model to a student by distilling from seemingly unrelated or random synthetic data from the teacher. This presents challenges in ensuring AI systems remain predictable and are trained safely, as standard auditing of the input data would not catch the hidden subliminal signal. Here, we investigate several open questions as to the enabling mechanisms and drivers of SL. First is the nature of the process by which biases are encoded in the data. We find that by adding Gaussian noise to the weights of the teacher and student models, the magnitude of subliminal transfer is increased by a factor of 1.9 in Gemma and 1.3 in Llama, suggesting that non-semantic weight structures play a crucial role. We show that steering vectors can be applied to the teacher to produce subliminal data, in addition to prompting and finetuning as used in previous studies. Analysis of the activations of the student models that have been trained on steered and prompted data demonstrates that students inherit not just the semantic meaning of the teacher's bias, but also the type of intervention that was used to apply it: steered students imitate steering vectors, prompted students do not. Additionally, the gradients of steered subliminal data show a linear correlation with the teacher's steering vectors, showing promise for data auditing. More broadly, as synthetic data becomes central to frontier training pipelines, being able to see the latent signals hidden in training data becomes paramount.

Can nonlinear dynamical systems be learned through a compact linear state-space representation, without directly solving a non-convex system-identification problem? We give a provable pipeline for doing so. Starting from observations of an unknown nonlinear dynamical system, we first learn an implicit spectral predictor using Observation Spectral Filtering (OSF), a convex method that competes with the best linear observer for the system. We then apply spectral-to-LDS distillation to convert this predictor into an explicit recurrent linear dynamical system. Our main theorem shows that the average prediction error of the distilled LDS decomposes into an exponentially-small distillation term and the OSF learning term governed by the Luenberger complexity of the best observer. The guarantee is dimension-free: it depends on observer complexity rather than on the latent dimension needed to represent the nonlinear system. To our knowledge, this yields the first end-to-end provable method for extracting a best-in-hindsight LDS representation of nonlinear dynamics through convex learning followed by provable distillation. Experiments on linear LDS benchmarks and MuJoCo behavior cloning show that the train-then-distill pipeline produces compact LDS predictors that match or outperform directly trained baselines.

Multi-turn agents trained with sparse outcome rewards receive limited supervision for intermediate decisions. Privileged on-policy self-distillation provides denser guidance by having a synchronized teacher re-score student responses with training-only references. However, the student may take different actions or complete subgoals in another order, reaching states for which the reference provides no applicable continuation. The resulting state--reference mismatch can lead the reference-conditioned teacher to penalize valid actions. We introduce State-Matched Routing and Contextualized Self-Distillation (SMRC-SD), which uses the student's reached state to decide when and how a reference should guide the student. At each turn, SMRC-SD checks state compatibility with the reference, distills only on matched turns, and contextualizes the reference by augmenting it with a reached-state summary and a compatible continuation. At inference, the policy requires neither references nor state matching. Experiments spanning three model sizes show gains over unconditional full-path self-distillation in embodied interaction (ALFWorld), web navigation (WebShop), and tool use (FTRL). FTRL-trained policies also transfer to BFCL and ToolHop without additional training. Code is available at https://github.com/liujunzhuo/SMRC-SD.

On-policy distillation (OPD) provides dense teacher supervision on student-generated trajectories, but standard reverse-KL training can assign insufficient probability to other plausible continuations. Teacher entropy alone does not reveal whether uncertainty is concentrated among a few plausible next tokens or dispersed over a long probability tail, nor whether the student already represents those candidates well. Moreover, local teacher probabilities may not predict downstream success. We introduce Sparse Probing and Outcome-calibrated Targets OPD (SPOT), which addresses two coupled decisions, where to probe and what to distill, through an acquisition--exploration--exploitation procedure. During acquisition, a position-level score combines normalized teacher entropy, the probability mass captured by a small top-$k$ candidate set, and student--teacher mismatch to allocate a limited probing budget. During exploration, SPOT evaluates teacher-proposed candidates through verifier-scored student continuations. During exploitation, these outcomes produce a closed-form, KL-regularized target that favors candidates with better downstream outcomes while remaining anchored to the teacher distribution. Extensive experiments across multiple student models and reasoning benchmarks demonstrate the effectiveness of SPOT in improving reasoning performance while balancing solution quality and coverage.

Small language models are often the only option for deployment under tight latency, cost, and on-premises constraints, but they are rarely trained from scratch: a compressed model is usually recovered through knowledge distillation (KD). This recovery step largely decides the final quality, yet it is expensive. We present a practitioner's study of how to make distillation training efficient, organised around two systems contributions. First, we show that offline KD (caching the teacher's top-$K$ logits once and training the student against the cache) matches online distillation at near-identical training loss while removing the teacher from memory, running about 29% faster per iteration, and reaching up to 41% higher throughput on a single H200 GPU. Second, we introduce a fused, chunked KL loss that never materialises the full vocabulary-sized logit tensor, making peak memory linear in the sequence length. This removes the memory spike that otherwise caps context length and lets us train at four times the context (32{,}768 tokens) on a single GPU. A separate output-head-only toy benchmark isolates the loss kernel and confirms its memory and iteration-rate scaling from 4K to 256K tokens. Together these make large-scale healing and hundreds of ablations affordable. We also report supporting ablations on loss design and sequence packing. We release our chunked-loss implementation: https://github.com/CompactifAI/Full-Chunked-KL-Loss.

Action labels tell a vision-language-action (VLA) policy which robot commands to imitate, but not how those commands change the 3D world. The aligned demonstration clip contains this missing supervision because its $K$ frame transitions record the geometry, motion, visibility, and camera change produced during the corresponding $K$ actions. We introduce Track4Action, a framework that distills this realized transition from a frozen world-centric 3D tracker into a current-observation VLA policy. During training, Track4World encodes the clip $V_{t:t+K}$ into a pooled tracker feature. Learnable track queries infer this feature from current VLA hidden states, match it in a shared space, and condition a flow-matching action head through a feature-wise gate. The tracker feature only defines the alignment target, so neither the clip nor the tracker is used at deployment. Track4Action reaches 82.3% on zero-shot LIBERO-Plus, improving the alignment-free variant by 7.6 points and LaMP by 3.0 points. It obtains 80.44% and 81.48% on the clean and randomized RoboTwin 2.0 splits, and 67.5% average success across four physical bimanual tasks, 25.0 points above the alignment-free variant. The gains across simulation and physical tasks support action-aligned 3D tracker features as privileged supervision for tracker-free VLA deployment. Our project page is available at https://wing0night.github.io/track4action-project-page.

On-Policy distillation (OPD) transfers teacher capabilities by supervising student-sampled trajectories with dense token-level teacher signals. Recent selective OPD methods improve this process by prioritizing signals that are confident, informative, or learnable. However, the assumptions overlook a fundamental failure mode of language models: their token-level judgments can be driven by input-agnostic language priors, formatting conventions, or stereotyped reasoning templates rather than task-specific evidence. We refer to such optimization-relevant but weakly input-grounded supervision as spurious signals in OPD, which may produce large gradients while contributing little task-improving direction. To mitigate this issue, we propose SA-OPD, a Spurious-Signal-Aware On-Policy Distillation framework that identifies and filters misleading token-level supervision based on input-groundedness and optimization impact. SA-OPD introduces a lightweight input-groundedness proxy estimating whether a token-level distillation signal truly depends on the input. It then filters only tokens that simultaneously exhibit low input-groundedness and extreme distillation divergence, thereby removing high-impact spurious updates and achieving fine-grained OPD optimization. Extensive experiments on both large language model (LLM) and vision-language model (VLM) settings demonstrate that SA-OPD consistently outperforms Vanilla OPD and competitive selective methods. These results establish input-groundedness as a key dimension for OPD supervision selection and offer a simple, effective strategy for mitigating spurious updates.

3D object detection using light detection and ranging (LiDAR) sensors requires a balance between accuracy and computational efficiency for onboard perception in autonomous driving and robotic navigation. Many existing LiDAR-based detection methods employ complex architectures to extract features, integrating large amounts of contextual information to enhance accuracy. This often results in significant computational costs, leading to suboptimal performance on resource-constrained embedded devices. In this study, we propose a knowledge distillation framework that transfers object-level voxel representations from a strong teacher model to lightweight student models through selective voxel-space feature alignment. Taking advantage of the linear-time sequence model with selective state spaces (Mamba), we design a multi-branch Mamba teacher backbone and a box-aware feature transfer mechanism that aligns spatially corresponding voxel features between teacher and student networks through a Mamba-based projection module. Experimental results on both a public dataset and real-world data show that our approach significantly reduces computational load while maintaining competitive accuracy compared with state-of-the-art methods.

Knowledge Distillation (KD) offers a promising yet underexplored path for compressing large action recognition models. However, existing KD methods suffer from two key limitations: 1) reliance on fixed input samples leads to suboptimal feature alignment between the frozen teacher (larger model) and the learnable student (smaller model), and 2) applying a uniform distillation strength for all channels fails to account for their varying importance in capturing distinct knowledge (e.g., motion tempo or magnitude) across training epochs. This motivates us to develop an Adaptive Sample-aware Channel-wise Dynamic (ASCD) KD approach, which operates in two stages. First, we use an adaptive sample generation module to create updated samples by incorporating semantics from sample gradients, which are derived by minimizing a feature loss weighted by channel centroid frequency differences at each layer. Meanwhile, crucial motion-related details are preserved by applying a Gaussian mask to frequency features. Second, we employ a channel-wise dynamic distillation module to train student on these generated samples, guided by sample gradients and feature frequencies. For efficiency, samples are updated periodically rather than per epoch. Extensive experiments on three video benchmarks (UCF101, Kinetics-400, Something-Something-v2) and two image datasets (CIFAR-100, ImageNet) demonstrate the state-of-the-art performance of our method. Code is available at https://github.com/mlvccn/ASCD_KD_Action.

On-policy self-distillation (OPSD), where a single model acts as both student and teacher with different contexts, has shown promise in verifiable domains like math, where hard privileged information (PI) in the form of ground-truth answers structurally constrains valid continuations. We extend OPSD to open-ended generation using soft PI in the form of rubrics that guide preferences but admit many valid responses. Rubrics have served as scalar rewards for reinforcement learning (RL); we show that they provide substantially richer signal as dense PI for distillation, and contrary to intuition, soft rubric PI provides a larger and more effective training signal on student roll-outs than hard reference completion PI in this regime. A reference completion is one point in a set of valid responses, so distilling towards it over-constrains the student, while rubrics specify the preference structure shared across the set of valid responses. We show the effectiveness of using rubrics as PI for open-ended generation across Qwen and Llama model families and show that it outperforms rubric-as-reward (RaR) RL using HealthBench, a benchmark that grades open-ended health responses against physician-created rubrics, providing dense token-level supervision for open-ended tasks; RuPI beats RaR RL by up to +0.10 absolute score and, under matched recipe and KL direction, beats reference-PI by +0.034 to +0.079 absolute score across three models. We further show that these findings generalize to training on the RubricHub Science corpus and evaluating on ResearchQA: soft rubric PI outperforms both reference-PI distillation and RaR RL (66.6% vs. 64.2% and 57.6%).