Knowledge Distillation

Ask about this topic

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

CardsList

829 papers164 papers in the last four weeks, up 95% on the four weeks before. 0.9% of all new papers.

Mar 28, 2026

From Foundation ECG Models to NISQ Learners: Distilling ECGFounder into a VQC Student

Foundation models have recently improved electrocardiogram (ECG) representation learning, but their deployment can be limited by computational cost and latency constraints. In this work, we fine-tune ECGFounder as a high-capacity teacher for binary ECG classification on PTB-XL and the MIT-BIH Arrhythmia Database, and investigate whether knowledge distillation can transfer its predictive behavior to compact students. We evaluate two classical 1D students (ResNet-1D and a lightweight CNN-1D) and a quantum-ready pipeline that combines a convolutional autoencoder, which compresses 256-sample ECG windows into a low-dimensional latent representation, with a 6-qubit variational quantum circuit implemented in Qiskit and executed in a simulated backend. Across both datasets, the teacher provides the strongest overall performance, while distillation yields competitive students under a considerable reduction in trainable parameters. We further analyze the sensitivity of student performance to distillation settings, highlighting consistent accuracy--efficiency trade-offs when compressing a foundation ECG model into classical and quantum-ready learners under a unified evaluation protocol.

Giovanni dos Santos Franco, Felipe Mahlow, Ellison Fernando Cardoso +1

Faculty of Sciences, S˜ao Paulo State University (Unesp), Bauru, Brazil. · Hospital Israelita Albert Einstein, S˜ao Paulo, Brazil. · Laboratory of Magnetic Resonance in Neuroradiology (LIM-44), University of S˜ao Paulo Faculty of Medicine Clinics Hospital, S˜ao Paulo, Brazil.

Electrocardiogram ClassificationKnowledge DistillationHybrid Quantum-Classical MLElectrocardiogram Foundation Models
AskPDF
Mar 27, 2026

Distilling Genomic Models for Efficient mRNA Representation Learning via Embedding Matching

Large Genomic Foundation Models have recently achieved remarkable results and in-vivo translation capabilities. However these models quickly grow to over a few Billion of parameters and are expensive to run when compute is limited. To overcome this challenge, we present a distillation framework for transferring mRNA representations from a state of the art genomic foundation model into a much smaller model specialized for mRNA sequences, reducing the size by 200-fold. Embedding-level distillation worked better than logit based methods, which we found unstable. Benchmarking on mRNA-bench demonstrates that the distilled model achieves state-of-the-art performance among models of comparable size and competes with larger architectures for mRNA-related tasks. Our results highlight embedding-based distillation of mRNA sequences as an effective training strategy for biological foundation models. This enables similar efficient and scalable sequence modelling in genomics, particularly when large models are computationally challenging or infeasible.

Rasched Haidari, Sam Martin, Maxime Allard

Helical London, UK

Knowledge DistillationMolecular Representation LearningFeature-Based Knowledge Distillation
AskPDF
Mar 27, 2026

GeoGuide: Hierarchical Geometric Guidance for Open-Vocabulary 3D Semantic Segmentation

Open-vocabulary 3D semantic segmentation aims to segment arbitrary categories beyond the training set. Existing methods predominantly rely on distilling knowledge from 2D open-vocabulary models. However, aligning 3D features to the 2D representation space restricts intrinsic 3D geometric learning and inherits errors from 2D predictions. To address these limitations, we propose GeoGuide, a novel framework that leverages pretrained 3D models to integrate hierarchical geometry-semantic consistency for open-vocabulary 3D segmentation. Specifically, we introduce an Uncertainty-based Superpoint Distillation module to fuse geometric and semantic features for estimating per-point uncertainty, adaptively weighting 2D features within superpoints to suppress noise while preserving discriminative information to enhance local semantic consistency. Furthermore, our Instance-level Mask Reconstruction module leverages geometric priors to enforce semantic consistency within instances by reconstructing complete instance masks. Additionally, our Inter-Instance Relation Consistency module aligns geometric and semantic similarity matrices to calibrate cross-instance consistency for same-category objects, mitigating viewpoint-induced semantic drift. Extensive experiments on ScanNet v2, Matterport3D, and nuScenes demonstrate the superior performance of GeoGuide.

Xujing Tao, Chuxin Wang, Yubo Ai +8

University of Science and Technology of China · National Key Laboratory of Deep Space Exploration, Deep Space Exploration Laboratory

Open-Vocabulary 3D Segmentation3D Semantic Segmentation3D Point Cloud SegmentationOpen-Vocabulary Semantic SegmentationKnowledge Distillation
AskPDF
Mar 27, 2026

Knowledge Distillation for Efficient Transformer-Based Reinforcement Learning in Hardware-Constrained Energy Management Systems

Transformer-based reinforcement learning has emerged as a strong candidate for sequential control in residential energy management. In particular, the Decision Transformer can learn effective battery dispatch policies from historical data, thereby increasing photovoltaic self-consumption and reducing electricity costs. However, transformer models are typically too computationally demanding for deployment on resource-constrained residential controllers, where memory and latency constraints are critical. This paper investigates knowledge distillation to transfer the decision-making behaviour of high-capacity Decision Transformer policies to compact models that are more suitable for embedded deployment. Using the Ausgrid dataset, we train teacher models in an offline sequence-based Decision Transformer framework on heterogeneous multi-building data. We then distil smaller student models by matching the teachers' actions, thereby preserving control quality while reducing model size. Across a broad set of teacher-student configurations, distillation largely preserves control performance and even yields small improvements of up to 1%, while reducing the parameter count by up to 96%, the inference memory by up to 90%, and the inference time by up to 63%. Beyond these compression effects, comparable cost improvements are also observed when distilling into a student model of identical architectural capacity. Overall, our results show that knowledge distillation makes Decision Transformer control more applicable for residential energy management on resource-limited hardware.

Pascal Henrich, Jonas Sievers, Maximilian Beichter +3

Karlsruhe Institute of Technology, Karlsruhe, Germany

Policy DistillationTransformer-Based RLKnowledge DistillationPower Systems
AskPDF
Mar 27, 2026

4DRaL: Bridging 4D Radar with LiDAR for Place Recognition using Knowledge Distillation

Place recognition is crucial for loop closure detection and global localization in robotics. Although mainstream algorithms typically rely on cameras and LiDAR, these sensors are susceptible to adverse weather conditions. Fortunately, the recently developed 4D millimeter-wave radar (4D radar) offers a promising solution for all-weather place recognition. However, the inherent noise and sparsity in 4D radar data significantly limit its performance. Thus, in this paper, we propose a novel framework called 4DRaL that leverages knowledge distillation (KD) to enhance the place recognition performance of 4D radar. Its core is to adopt a high-performance LiDAR-to-LiDAR (L2L) place recognition model as a teacher to guide the training of a 4D radar-to-4D radar (R2R) place recognition model. 4DRaL comprises three key KD modules: a local image enhancement module to handle the sparsity of raw 4D radar points, a feature distribution distillation module that ensures the student model generates more discriminative features, and a response distillation module to maintain consistency in feature space between the teacher and student models. More importantly, 4DRaL can also be trained for 4D radar-to-LiDAR (R2L) place recognition through different module configurations. Experimental results prove that 4DRaL achieves state-of-the-art performance in both R2R and R2L tasks regardless of normal or adverse weather.

Ningyuan Huang, Zhiheng Li, Zheng Fang

The authors are all with the Faculty of Robot Science and Engineering, Northeastern University, Shenyang, China.

Millimeter-Wave RadarKnowledge Distillation
AskPDF
Mar 27, 2026

HAD: Heterogeneity-Aware Distillation for Lifelong Heterogeneous Learning

Lifelong learning aims to preserve knowledge acquired from previous tasks while incorporating knowledge from a sequence of new tasks. However, most prior work explores only streams of homogeneous tasks (e.g., only classification tasks) and neglects the scenario of learning across heterogeneous tasks that possess different structures of outputs. In this work, we formalize this broader setting as lifelong heterogeneous learning (LHL). Departing from conventional lifelong learning, the task sequence of LHL spans different task types, and the learner needs to retain heterogeneous knowledge for different output space structures. To instantiate the LHL, we focus on LHL in the context of dense prediction (LHL4DP), a realistic and challenging scenario. To this end, we propose the Heterogeneity-Aware Distillation (HAD) method, an exemplar-free approach that preserves previously gained heterogeneous knowledge by self-distillation in each training phase. The proposed HAD comprises two complementary components, including a distribution-balanced heterogeneity-aware distillation loss to alleviate the global imbalance of prediction distribution and a salience-guided heterogeneity-aware distillation loss that concentrates learning on informative edge pixels extracted with the Sobel operator. Extensive experiments demonstrate that the proposed HAD method significantly outperforms existing methods in this new scenario.

Xuerui Zhang, Xuehao Wang, Zhan Zhuang +5

Southern University of Science and Technology · Technical University of Munich · Zhejiang University +2

Knowledge DistillationDense Prediction
AskPDF
Mar 27, 2026

Dual-Stage Invariant Continual Learning under Extreme Visual Sparsity

Continual learning seeks to maintain stable adaptation under non-stationary environments, yet this problem becomes particularly challenging in object detection, where most existing methods implicitly assume relatively balanced visual conditions. In extreme-sparsity regimes, such as those observed in space-based resident space object (RSO) detection scenarios, foreground signals are overwhelmingly dominated by background observations. Under such conditions, we analytically demonstrate that background-driven gradients destabilize the feature backbone during sequential domain shifts, causing progressive representation drift. This exposes a structural limitation of continual learning approaches relying solely on output-level distillation, as they fail to preserve intermediate representation stability. To address this, we propose a dual-stage invariant continual learning framework via joint distillation, enforcing structural and semantic consistency on both backbone representations and detection predictions, respectively, thereby suppressing error propagation at its source while maintaining adaptability. Furthermore, to regulate gradient statistics under severe imbalance, we introduce a sparsity-aware data conditioning strategy combining patch-based sampling and distribution-aware augmentation. Experiments on a high-resolution space-based RSO detection dataset show consistent improvement over established continual object detection methods, achieving an absolute gain of +4.0 mAP under sequential domain shifts.

Rangya Zhang, Jiaping Xiao, Lu Bai +2

School of Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore 639798, Singapore

Continual Object DetectionIncremental LearningObject DetectionKnowledge Distillation
AskPDF
Mar 27, 2026

Efficient Few-Shot Learning for Edge AI via Knowledge Distillation on MobileViT

Efficient and adaptable deep learning models are an important area of deep learning research, driven by the need for highly efficient models on edge devices. Few-shot learning enables the use of deep learning models in low-data regimes, a capability that is highly sought after in real-world applications where collecting large annotated datasets is costly or impractical. This challenge is particularly relevant in edge scenarios, where connectivity may be limited, low-latency responses are required, or energy consumption constraints are critical. We propose and evaluate a pre-training method for the MobileViT backbone designed for edge computing. Specifically, we employ knowledge distillation, which transfers the generalization ability of a large-scale teacher model to a lightweight student model. This method achieves accuracy improvements of 14% and 6.7% for one-shot and five-shot classification, respectively, on the MiniImageNet benchmark, compared to the ResNet12 baseline, while reducing by 69% the number of parameters and by 88% the computational complexity of the model, in FLOPs. Furthermore, we deployed the proposed models on a Jetson Orin Nano platform and measured power consumption directly at the power supply, showing that the dynamic energy consumption is reduced by 37% with a latency of 2.6 ms. These results demonstrate that the proposed method is a promising and practical solution for deploying few-shot learning models on edge AI hardware.

Shuhei Tsuyuki, Reda Bensaid, Jérémy Morlier +4

Research Institute of Electrical Communication, Tohoku University, Japan · Graduate School of Engineering, Tohoku University, Japan · IMT Atlantique, Lab-STICC, UMR CNRS 6285, F-29238 Brest, France

Few-Shot Image ClassificationEdge ComputingEnergy-Efficient MLKnowledge Distillation
AskPDF
Mar 27, 2026

Learnable Instance Attention Filtering for Adaptive Detector Distillation

As deep vision models grow increasingly complex to achieve higher performance, deployment efficiency has become a critical concern. Knowledge distillation (KD) mitigates this issue by transferring knowledge from large teacher models to compact student models. While many feature-based KD methods rely on spatial filtering to guide distillation, they typically treat all object instances uniformly, ignoring instance-level variability. Moreover, existing attention filtering mechanisms are typically heuristic or teacher-driven, rather than learned with the student. To address these limitations, we propose Learnable Instance Attention Filtering for Adaptive Detector Distillation (LIAF-KD), a novel framework that introduces learnable instance selectors to dynamically evaluate and reweight instance importance during distillation. Notably, the student contributes to this process based on its evolving learning state. Experiments on the KITTI and COCO datasets demonstrate consistent improvements, with a 2% gain on a GFL ResNet-50 student without added complexity, outperforming state-of-the-art methods.

Chen Liu, Qizhen Lan, Zhicheng Ding +2

Dept. of CS, University of Alabama at Birmingham · Dept. of CS, Bowling Green State University

Knowledge DistillationFeature-Based Knowledge DistillationSelective Knowledge DistillationEfficient Object Detection
AskPDF
Mar 26, 2026

Training the Knowledge Base through Evidence Distillation and Write-Back Enrichment

The knowledge base in a retrieval-augmented generation (RAG) system is typically assembled once and never revised, even though the facts a query requires are often fragmented across documents and buried in irrelevant content. We argue that the knowledge base should be treated as a trainable component and propose WriteBack-RAG, a framework that uses labeled examples to identify where retrieval succeeds, isolate the relevant documents, and distill them into compact knowledge units that are indexed alongside the original corpus. Because the method modifies only the corpus, it can be applied once as an offline preprocessing step and combined with any RAG pipeline. Across four RAG methods, six benchmarks, and two LLM backbones, WriteBack-RAG improves every evaluated setting, with gains averaging +2.14%. Cross-method transfer experiments further show that the distilled knowledge benefits RAG pipelines other than the one used to produce it, confirming that the improvement resides in the corpus itself.

Yuxing Lu, Xukai Zhao, Wei Wu +1

Peking University · Georgia Institute of Technology · Tsinghua University

Retrieval-Augmented GenerationKnowledge Augmentation for Language ModelsKnowledge Distillation
AskPDF
Mar 26, 2026

Neural Network Conversion of Machine Learning Pipelines

Transfer learning and knowledge distillation has recently gained a lot of attention in the deep learning community. One transfer approach, the student-teacher learning, has been shown to successfully create small'' student neural networks that mimic the performance of a much bigger and more complex teacher'' networks. In this paper, we investigate an extension to this approach and transfer from a non-neural-based machine learning pipeline as teacher to a neural network (NN) student, which would allow for joint optimization of the various pipeline components and a single unified inference engine for multiple ML tasks. In particular, we explore replacing the random forest classifier by transfer learning to a student NN. We experimented with various NN topologies on 100 OpenML tasks in which random forest has been one of the best solutions. Our results show that for the majority of the tasks, the student NN can indeed mimic the teacher if one can select the right NN hyper-parameters. We also investigated the use of random forest for selecting the right NN hyper-parameters.

Man-Ling Sung, Jan Silovsky, Man-Hung Siu +2

Raytheon BBN Technologies, 10 Moulton Street, Cambridge, MA 02138

Transfer LearningKnowledge Distillation
AskPDF
Mar 26, 2026

Residual-as-Teacher: Mitigating Bias Propagation in Student--Teacher Estimation

We study statistical estimation in a student--teacher setting, where predictions from a pre-trained teacher are used to guide a student model. A standard approach is to train the student to directly match the teacher's outputs, which we refer to as student soft matching (SM). This approach directly propagates any systematic bias or mis-specification present in the teacher, thereby degrading the student's predictions. We propose and analyze an alternative scheme, known as residual-as-teacher (RaT), in which the teacher is used to estimate residuals in the student's predictions. Our analysis shows how the student can thereby emulate a proximal gradient scheme for solving an oracle optimization problem, and this provably reduces the effect of teacher bias. For general student--teacher pairs, we establish non-asymptotic excess risk bounds for any RaT fixed point, along with convergence guarantees for the student-teacher iterative scheme. For kernel-based student--teacher pairs, we prove a sharp separation: the RaT method achieves the minimax-optimal rate, while the SM method incurs constant prediction error for any sample size. Experiments on both synthetic data and ImageNette classification under covariate shift corroborate our theoretical findings.

Kakei Yamamoto, Martin J. Wainwright

Lab for Information and Decision Systems; Statistics and Data Science Center; EECS; Massachusetts Institute of Technology · Lab for Information and Decision Systems; Statistics and Data Science Center; EECS and Mathematics; Massachusetts Institute of Technology

Statistical Learning TheoryKnowledge Distillation
AskPDF
Mar 26, 2026

Multimodal Dataset Distillation via Phased Teacher Models

Multimodal dataset distillation aims to construct compact synthetic datasets that enable efficient compression and knowledge transfer from large-scale image-text data. However, existing approaches often fail to capture the complex, dynamically evolving knowledge embedded in the later training stages of teacher models. This limitation leads to degraded student performance and compromises the quality of the distilled data. To address critical challenges such as pronounced cross-stage performance gaps and unstable teacher trajectories, we propose Phased Teacher Model with Shortcut Trajectory (PTM-ST) -- a novel phased distillation framework. PTM-ST leverages stage-aware teacher modeling and a shortcut-based trajectory construction strategy to accurately fit the teacher's learning dynamics across distinct training phases. This enhances both the stability and expressiveness of the distillation process. Through theoretical analysis and comprehensive experiments, we show that PTM-ST significantly mitigates optimization oscillations and inter-phase knowledge gaps, while also reducing storage overhead. Our method consistently surpasses state-of-the-art baselines on Flickr30k and COCO, achieving up to 13.5% absolute improvement and an average gain of 9.53% on Flickr30k. Code: https://github.com/Previsior/PTM-ST.

Shengbin Guo, Hang Zhao, Senqiao Yang +5

Harbin Institute of Technology, Shenzhen · The Chinese University of Hong Kong

Dataset DistillationKnowledge Distillation
AskPDF
Mar 26, 2026

CLIP-RD: Relational Distillation for Efficient CLIP Knowledge Distillation

Contrastive Language-Image Pre-training (CLIP) demonstrates strong zero-shot generalization, but due to substantial computational and memory costs, distillation into lightweight models is required. Existing relational objectives do not explicitly model multidirectional relationships between teacher and student embeddings, potentially leaving the geometric relationships insufficiently constrained. This may disrupt the modality-gap structure important for zero-shot transfer. To address these limitations, we propose a relational distillation framework, CLIP-RD, which introduces two relational methods, Vertical Relational Distillation (VRD) and Cross Relational Distillation (XRD). VRD aligns teacher-student intra-modal similarity distributions to enforce consistent distillation strength across image and text embeddings. Meanwhile, XRD aligns the teacher-image-student-text and teacher-text-student-image similarity distributions to impose bidirectional cross-modal symmetry. By jointly modeling these multidirectional relational structures, CLIP-RD aligns the student's embedding geometry more faithfully to the teacher's, outperforming CLIP-KD by 1.8%p. This performance improvement is maintained across diverse architectures, teacher scales, retrieval tasks, downstream tasks, and corruption settings, with negligible additional training-time overhead.

Jeannie Chung, Hanna Jang, Ingyeong Yang +2

Ewha Womans University, Seoul 03760, South Korea

Cross-Modal LearningCross-Modal Knowledge DistillationVision-Language ModelsVLM DistillationRelational Knowledge DistillationKnowledge Distillation
AskPDF
Mar 26, 2026

A Public Theory of Distillation Resistance via Constraint-Coupled Reasoning Architectures

Knowledge distillation, model extraction, and behavior transfer have become central concerns in frontier AI. The main risk is not merely copying, but the possibility that useful capability can be transferred more cheaply than the governance structure that originally accompanied it. This paper presents a public, trade-secret-safe theoretical framework for reducing that asymmetry at the architectural level. The core claim is that distillation becomes less valuable as a shortcut when high-level capability is coupled to internal stability constraints that shape state transitions over time. To formalize this idea, the paper introduces a constraint-coupled reasoning framework with four elements: bounded transition burden, path-load accumulation, dynamically evolving feasible regions, and a capability-stability coupling condition. The paper is intentionally public-safe: it omits proprietary implementation details, training recipes, thresholds, hidden-state instrumentation, deployment procedures, and confidential system design choices. The contribution is therefore theoretical rather than operational. It offers a falsifiable architectural thesis, a clear threat model, and a set of experimentally testable hypotheses for future work on distillation resistance, alignment, and model governance.

Peng Wei, Wesley Shu

College of Plant Protection, Southwest University, China · The Institute of Energetic Paradigm

Knowledge Distillation
AskPDF
Mar 25, 2026

Powerful Teachers Matter: Text-Guided Multi-view Knowledge Distillation with Visual Prior Enhancement

Knowledge distillation transfers knowledge from large teacher models to smaller students for efficient inference. While existing methods primarily focus on distillation strategies, they often overlook the importance of enhancing teacher knowledge quality. In this paper, we propose Text-guided Multi-view Knowledge Distillation (TMKD), which leverages dual-modality teachers, a visual teacher and a text teacher (CLIP), to provide richer supervisory signals. Specifically, we enhance the visual teacher with multi-view inputs incorporating visual priors (edge and high-frequency features), while the text teacher generates semantic weights through prior-aware prompts to guide adaptive feature fusion. Additionally, we introduce vision-language contrastive regularization to strengthen semantic knowledge in the student model. Extensive experiments on five benchmarks demonstrate that TMKD consistently improves knowledge distillation performance by up to 4.49%, validating the effectiveness of our dual-teacher multi-view enhancement strategy. Code is available at https://anonymous.4open.science/r/TMKD-main-44D1.

Xin Zhang, Jianyang Xu, Hao Peng +5

Hangzhou Dianzi University

Knowledge DistillationMulti-Teacher Knowledge DistillationFeature-Based Knowledge DistillationCross-Modal Knowledge Distillation
AskPDF
Mar 25, 2026

CAKE: Real-time Action Detection via Motion Distillation and Background-aware Contrastive Learning

Online Action Detection (OAD) systems face two primary challenges: high computational cost and insufficient modeling of discriminative temporal dynamics against background motion. Adding optical flow could provides strong motion cues but it incurs significant computational overhead. We propose CAKE, a OAD Flow-based distillation framework to transfer motion knowledge into RGB models. We propose Dynamic Motion Adapter (DMA) to suppress static background noise and emphasize pixel changes, effectively approximating optical flow without explicit computation. The framework also integrates a Floating Contrastive Learning strategy to distinguish informative motion dynamics from temporal background. Various experiments conducted on the TVSeries, THUMOS'14, Kinetics-400 datasets show effectiveness of our model. CAKE achieves a standout mAP compared with SOTA while using the same backbone. Our model operates at over 72 FPS on a single CPU, making it highly suitable for resource-constrained systems.

Hieu Hoang, Dung Trung Tran, Hong Nguyen +1

Viettel Group, Vietnam · School of Electrical and Electronics Engineering, Hanoi University of Science and Technology, Vietnam · Signal Analysis and Interpretation Lab, University of Southern California, United States

Temporal Action DetectionHuman Activity RecognitionKnowledge Distillation
AskPDF
Mar 25, 2026

TED: Training-Free Experience Distillation for Multimodal Reasoning

Knowledge distillation is typically realized by transferring a teacher model's knowledge into a student's parameters through supervised or reinforcement-based optimization. While effective, such approaches require repeated parameter updates and large-scale training data, limiting their applicability in resource-constrained environments. In this work, we propose TED, a training-free, context-based distillation framework that shifts the update target of distillation from model parameters to an in-context experience injected into the student's prompt. For each input, the student generates multiple reasoning trajectories, while a teacher independently produces its own solution. The teacher then compares the student trajectories with its reasoning and the ground-truth answer, extracting generalized experiences that capture effective reasoning patterns. These experiences are continuously refined and updated over time. A key challenge of context-based distillation is unbounded experience growth and noise accumulation. TED addresses this with an experience compression mechanism that tracks usage statistics and selectively merges, rewrites, or removes low-utility experiences. Experiments on multimodal reasoning benchmarks MathVision and VisualPuzzles show that TED consistently improves performance. On MathVision, TED raises the performance of Qwen3-VL-8B from 0.627 to 0.702, and on VisualPuzzles from 0.517 to 0.561 with just 100 training samples. Under this low-data, no-update setting, TED achieves performance competitive with fully trained parameter-based distillation while reducing training cost by over 5x, demonstrating that meaningful knowledge transfer can be achieved through contextual experience.

Shuozhi Yuan, Jinqing Wang, Zihao Liu +5

China Telecom Digital Intelligence Technology Co.,Ltd., Beijing, China · Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China

Context DistillationKnowledge DistillationMultimodal Reasoning
AskPDF
Mar 24, 2026

MemCollab: Cross-Model Memory Collaboration via Contrastive Trajectory Distillation

LLM agents increasingly rely on memory mechanisms to reuse knowledge from past problem-solving experiences. However, existing methods typically construct memory for a single agent and reuse it with the same underlying model, tightly coupling stored knowledge to model-specific reasoning styles. In heterogeneous deployments, where agents may be instantiated with backbone models of different sizes, architectures, or specializations, this raises a key question: can a single memory system be shared across agents with different backbone models? We find that naive cross-model memory transfer can degrade performance, because stored memories often entangle task-relevant knowledge with model-specific biases. To address this challenge, we propose MemCollab, a collaborative memory framework that builds shared cross-model memory by contrasting reasoning trajectories generated by different model-based agents on the same task. Through this contrastive process, MemCollab distills abstract reasoning constraints that capture shared task-level invariants while suppressing model-specific artifacts. We further introduce a task-aware retrieval mechanism that conditions memory access on task category, ensuring that only relevant constraints are retrieved at inference time. Experiments on mathematical reasoning and code generation benchmarks show that MemCollab consistently improves both accuracy and inference-time efficiency across diverse agents, including settings with different model families. These results demonstrate that collaboratively constructed cross-model memory can serve as a shared reasoning resource for heterogeneous LLM-based agents.

Yurui Chang, Yiran Wu, Qingyun Wu +1

Pennsylvania State University · AG2AI

Memory-Augmented Language Model AgentsMulti-Agent CollaborationKnowledge Distillation
AskPDF
Mar 24, 2026

Adaptive Dual-Teacher Distillation with Subnetwork Rectification for Bridging Semantic Gaps in Black-Box Domain Adaptation

Assuming that neither source data nor source model parameters are accessible, black-box domain adaptation (BBDA) represents a highly practical yet challenging setting, where transferable knowledge is limited to the predictions of a black-box source model. Existing approaches exploit such knowledge via pseudo-label refinement or by leveraging vision-language models (ViLs), but they often fail to reconcile the inherent discrepancy between task-specific knowledge from black-box models and language-aligned semantic priors of ViLs, resulting in suboptimal integration and degraded adaptation performance. To address this challenge, we propose adaptive Dual-Teacher Distillation with Subnetwork Rectification (DDSR), a framework that explicitly reconciles these complementary yet inconsistent knowledge sources. DDSR employs an adaptive prediction fusion strategy to integrate predictions from the black-box source model and a ViL, generating reliable pseudo-labels for the target domain. A subnetwork-based regularization mechanism mitigates overfitting to noisy supervision by enforcing output consistency and gradient divergency. Furthermore, progressively improved target predictions iteratively refine both pseudo-labels and ViL prompts, enhancing semantic alignment. Finally, class-wise prototypes are used to further optimize target predictions via self-training. Extensive experiments on multiple benchmark datasets demonstrate that DDSR consistently outperforms state-of-the-art methods, including those with access to source data or source model parameters.

Zhe Zhang, Jing Li, Wanli Xue +4

School of Computer Science and Engineering, Tianjin University of Technology, Tianjin 300384, China. · School of Artificial Intelligence, Tianjin University, Tianjin 300350, China

Source-Free Domain AdaptationKnowledge DistillationVLM DistillationVLM Adaptation
AskPDF
Mar 24, 2026

Transformers Trained via Gradient Descent Can Provably Learn a Class of Teacher Models

Transformers have achieved great success across a wide range of applications, yet the theoretical foundations underlying their success remain largely unexplored. To demystify the strong capacities of transformers applied to versatile scenarios and tasks, we theoretically investigate utilizing transformers as students to learn from a class of teacher models. Specifically, the teacher models covered in our analysis include convolution layers with average pooling, graph convolution layers, and various classic statistical learning models, including a variant of sparse token selection models [Sanford et al., 2023, Wang et al., 2024] and group-sparse linear predictors [Zhang et al., 2025]. When learning from this class of teacher models, we prove that one-layer transformers with simplified "position-only'' attention can successfully recover all parameter blocks of the teacher models, thus achieving the optimal population loss. Building upon the efficient mimicry of trained transformers towards teacher models, we further demonstrate that they can generalize well to a broad class of out-of-distribution data under mild assumptions. The key in our analysis is to identify a fundamental bilinear structure shared by various learning tasks, which enables us to establish unified learning guarantees for these tasks when treating them as teachers for transformers.

Chenyang Zhang, Qingyue Zhao, Quanquan Gu +1

The University of Hong Kong · University of California, Los Angeles

TransformerStatistical Learning TheoryKnowledge Distillation
AskPDF
Mar 23, 2026

Clinical Graph-Mediated Distillation for Unpaired MRI-to-CFI Hypertension Prediction

Retinal fundus imaging enables low-cost and scalable hypertension (HTN) screening, but HTN-related retinal cues are subtle, yielding high-variance predictions. Brain MRI provides stronger vascular and small-vessel-disease markers of HTN, yet it is expensive and rarely acquired alongside fundus images, resulting in modality-siloed datasets with disjoint MRI and fundus cohorts. We study this unpaired MRI-fundus regime and introduce Clinical Graph-Mediated Distillation (CGMD), a framework that transfers MRI-derived HTN knowledge to a fundus model without paired multimodal data. CGMD leverages shared structured biomarkers as a bridge by constructing a clinical similarity kNN graph spanning both cohorts. We train an MRI teacher, propagate its representations over the graph, and impute brain-informed representation targets for fundus patients. A fundus student is then trained with a joint objective combining HTN supervision, target distillation, and relational distillation. Experiments on our newly collected unpaired MRI-fundus-biomarker dataset show that CGMD consistently improves fundus-based HTN prediction over standard distillation and non-graph imputation baselines, with ablations confirming the importance of clinically grounded graph connectivity. Code is available at https://github.com/DillanImans/CGMD-unpaired-distillation.

Dillan Imans, Phuoc-Nguyen Bui, Duc-Tai Le +1

Department of Computer Science and Engineering, Sungkyunkwan University, Suwon, South Korea · Convergence Research Institute, Sungkyunkwan University, Suwon, South Korea · Department of AI Systems Engineering, Sungkyunkwan University, Suwon, South Korea +1

Cross-Modal Knowledge DistillationClinical PredictionKnowledge DistillationFundus Imaging
AskPDF
Mar 22, 2026

Uncertainty-Aware Knowledge Distillation for Multimodal Large Language Models

Knowledge distillation establishes a learning paradigm that leverages both data supervision and teacher guidance. However, determining the optimal balance between learning from data and learning from the teacher is challenging, as some samples may be noisy while others are subject to teacher uncertainty. This motivates the need for adaptively balancing data and teacher supervision. We propose Beta-weighted Knowledge Distillation (Beta-KD), an uncertainty-aware distillation framework that adaptively modulates how much the student relies on teacher guidance. Specifically, we formulate teacher--student learning from a unified Bayesian perspective and interpret teacher supervision as a Gibbs prior over student activations. This yields a closed-form, uncertainty-aware weighting mechanism and supports arbitrary distillation objectives and their combinations. Extensive experiments on multimodal VQA benchmarks demonstrate that distilling student Vision-Language Models from a large teacher VLM consistently improves performance. The results show that Beta-KD outperforms existing knowledge distillation methods. The code is available at https://github.com/Jingchensun/beta-kd.

Jingchen Sun, Shaobo Han, Deep Patel +3

NEC Laboratories America, Inc., USA · University at Buffalo, SUNY · Rutgers University

Uncertainty-Aware Knowledge DistillationKnowledge DistillationVLM DistillationMultimodal Large Language Models
AskPDF
Mar 22, 2026

Demystifying Low-Rank Knowledge Distillation in Large Language Models: Convergence, Generalization, and Information-Theoretic Guarantees

Knowledge distillation has emerged as a powerful technique for compressing large language models (LLMs) into efficient, deployable architectures while preserving their advanced capabilities. Recent advances in low-rank knowledge distillation, particularly methods like Low-Rank Clone (LRC), have demonstrated remarkable empirical success, achieving comparable performance to full-parameter distillation with significantly reduced training data and computational overhead. However, the theoretical foundations underlying these methods remain poorly understood. In this paper, we establish a rigorous theoretical framework for low-rank knowledge distillation in language models. We prove that under mild assumptions, low-rank projection preserves the optimization dynamics, yielding explicit convergence rates of O(1/T)O(1/\sqrt{T}). We derive generalization bounds that characterize the fundamental trade-off between model compression and generalization capability, showing that the generalization error scales with the rank parameter as O(r(m+n)/n)O(r(m+n)/\sqrt{n}). Furthermore, we provide an information-theoretic analysis of the activation cloning mechanism, revealing its role in maximizing the mutual information between the teacher's and student's intermediate representations. Our theoretical results offer principled guidelines for rank selection, mathematically suggesting an optimal rank r∗=O(n)r^* = O(\sqrt{n}) where nn is the sample size. Experimental validation on standard language modeling benchmarks confirms our theoretical predictions, demonstrating that the empirical convergence, rank scaling, and generalization behaviors align closely with our bounds.

Alberlucia Rafael Soarez, Daniel Kim, Mariana Costa +1

Department of Computer Science, University of Brasilia

Language Model DistillationKnowledge Distillation
AskPDF
Mar 21, 2026

VSD-MOT: End-to-End Multi-Object Tracking in Low-Quality Video Scenes Guided by Visual Semantic Distillation

Existing multi-object tracking algorithms typically fail to adequately address the issues in low-quality videos, resulting in a significant decline in tracking performance when image quality deteriorates in real-world scenarios. This performance degradation is primarily due to the algorithms' inability to effectively tackle the problems caused by information loss in low-quality images. To address the challenges of low-quality video scenarios, inspired by vision-language models, we propose a multi-object tracking framework guided by visual semantic distillation (VSD-MOT). Specifically, we introduce the CLIP Image Encoder to extract global visual semantic information from images to compensate for the loss of information in low-quality images. However, direct integration can substantially impact the efficiency of the multi-object tracking algorithm. Therefore, this paper proposes to extract visual semantic information from images through knowledge distillation. This method adopts a teacher-student learning framework, with the CLIP Image Encoder serving as the teacher model. To enable the student model to acquire the capability of extracting visual semantic information suitable for multi-object tracking tasks from the teacher model, we have designed the Dual-Constraint Semantic Distillation method (DCSD). Furthermore, to address the dynamic variation of frame quality in low-quality videos, we propose the Dynamic Semantic Weight Regulation (DSWR) module, which adaptively allocates fusion weights based on real-time frame quality assessment. Extensive experiments demonstrate the effectiveness and superiority of the proposed method in low-quality video scenarios in the real world. Meanwhile, our method can maintain good performance in conventional scenarios.

Jun Du

Multi-Object TrackingVisual Object TrackingKnowledge DistillationVLM Distillation
AskPDF
Mar 19, 2026

EdgeCrafter: Compact ViTs for Edge Dense Prediction via Task-Specialized Distillation

Deploying high-performance dense prediction models on resource-constrained edge devices remains challenging due to strict computation and memory budgets. In practice, lightweight systems for object detection, instance segmentation, and pose estimation are still dominated by CNN-based architectures such as YOLO, while compact Vision Transformers (ViTs) often struggle to achieve comparable accuracy-efficiency trade-offs, even with large-scale pretraining. We argue that this gap arises primarily from insufficient task-specific representation learning in small-scale ViTs, rather than from an inherent mismatch between ViTs and edge dense prediction. To address this issue, we introduce EdgeCrafter, a unified compact ViT framework for edge dense prediction centered on ECDet, a detection model built on a distilled compact backbone and an edge-friendly encoder-decoder design. The resulting detection-distilled representation transfers directly to instance segmentation and human pose estimation through lightweight task-specific prediction modules. Without using task annotations beyond COCO, ECDet-S achieves 51.7 box AP with fewer than 10M parameters, while ECInsSeg-X and ECPose-X reach 48.4 mask AP and 74.8 keypoint AP, respectively. As a complementary but more compute-intensive setting, Objects365 detection pretraining consistently improves performance across all scales, with the X variants reaching 59.9 box AP, 49.8 mask AP, and 75.9 keypoint AP. These results show that compact ViTs, when combined with task-specialized distillation and edge-aware design, can be a practical and competitive solution for edge dense prediction. Code is available at: https://intellindust-ai-lab.github.io/projects/EdgeCrafter

Longfei Liu, Yongjie Hou, Yang Li +7

Intellindust AI Lab

Dense PredictionEfficient ViTsEfficient Object DetectionEdge InferenceInstance SegmentationKnowledge Distillation
AskPDF
Mar 19, 2026

Collaborative Adaptive Curriculum for Progressive Knowledge Distillation

Recent advances in collaborative knowledge distillation have demonstrated cutting-edge performance for resource-constrained distributed multimedia learning scenarios. However, achieving such competitiveness requires addressing a fundamental mismatch: high-dimensional teacher knowledge complexity versus heterogeneous client learning capacities, which currently prohibits deployment in edge-based visual analytics systems. Drawing inspiration from curriculum learning principles, we introduce Federated Adaptive Progressive Distillation (FAPD), a consensus-driven framework that orchestrates adaptive knowledge transfer. FAPD hierarchically decomposes teacher features via PCA-based structuring, extracting principal components ordered by variance contribution to establish a natural visual knowledge hierarchy. Clients progressively receive knowledge of increasing complexity through dimension-adaptive projection matrices. Meanwhile, the server monitors network-wide learning stability by tracking global accuracy fluctuations across a temporal consensus window, advancing curriculum dimensionality only when collective consensus emerges. Consequently, FAPD provably adapts knowledge transfer pace while achieving superior convergence over fixed-complexity approaches. Extensive experiments on three datasets validate FAPD's effectiveness: it attains 3.64% accuracy improvement over FedAvg on CIFAR-10, demonstrates 2x faster convergence, and maintains robust performance under extreme data heterogeneity (α=0.1), outperforming baselines by over 4.5%.

Jing Liu, Zhenchao Ma, Han Yu +5

Fudan University · The University of British Columbia · Duke Kunshan University +2

Federated Knowledge DistillationKnowledge DistillationFeature-Based Knowledge DistillationHeterogeneous Federated LearningCurriculum Learning
AskPDF
Mar 19, 2026

HSI Image Enhancement Classification Based on Knowledge Distillation: A Study on Forgetting

In incremental classification tasks for hyperspectral images, catastrophic forgetting is an unavoidable challenge. While memory recall methods can mitigate this issue, they heavily rely on samples from old categories. This paper proposes a teacher-based knowledge retention method for incremental image classification. It alleviates model forgetting of old category samples by utilizing incremental category samples, without depending on old category samples. Additionally, this paper introduces a mask-based partial category knowledge distillation algorithm. By decoupling knowledge distillation, this approach filters out potentially misleading information that could misguide the student model, thereby enhancing overall accuracy. Comparative and ablation experiments demonstrate the proposed method's robust performance.

Songfeng Zhu

School of Software Engineering, Henan Polytechnic University, Henan 454000, china

Class-Incremental LearningHyperspectral Image ClassificationCatastrophic ForgettingKnowledge DistillationSelective Knowledge Distillation
AskPDF
Mar 17, 2026

Integrating Inductive Biases in Transformers via Distillation for Financial Time Series Forecasting

Transformer-based models have been widely adopted for time-series forecasting due to their high representational capacity and architectural flexibility. However, many Transformer variants implicitly assume stationarity and stable temporal dynamics -- assumptions routinely violated in financial markets characterized by regime shifts and non-stationarity. Empirically, state-of-the-art time-series Transformers often underperform even vanilla Transformers on financial tasks, while simpler architectures with distinct inductive biases, such as CNNs and RNNs, can achieve stronger performance with substantially lower complexity. At the same time, no single inductive bias dominates across markets or regimes, suggesting that robust financial forecasting requires integrating complementary temporal priors. We propose TIPS (Transformer with Inductive Prior Synthesis), a knowledge distillation framework that synthesizes diverse inductive biases -- causality, locality, and periodicity -- within a unified Transformer. TIPS trains bias-specialized Transformer teachers via attention masking, then distills their knowledge into a single student model with regime-dependent alignment across inductive biases. Across four major equity markets, TIPS achieves state-of-the-art performance, outperforming strong ensemble baselines by 55%, 9%, and 16% in annual return, Sharpe ratio, and Calmar ratio, while requiring only 38% of the inference-time computation. Further analyses show that TIPS generates statistically significant excess returns beyond both vanilla Transformers and its teacher ensembles, and exhibits regime-dependent behavioral alignment with classical architectures during their profitable periods. These results highlight the importance of regime-dependent inductive bias utilization for robust generalization in non-stationary financial time series.

Yu-Chen Den, Kuan-Yu Chen, Kendro Vincent +1

SinoPac Holdings, Taipei, Taiwan · National Chengchi University, Taipei, Taiwan

Time Series ForecastingFinancial Time SeriesTime Series TransformersNon-Stationary Time Series ForecastingKnowledge Distillation
AskPDF
Mar 17, 2026

CD-FKD: Cross-Domain Feature Knowledge Distillation for Robust Single-Domain Generalization in Object Detection

Single-domain generalization is essential for object detection, particularly when training models on a single source domain and evaluating them on unseen target domains. Domain shifts, such as changes in weather, lighting, or scene conditions, pose significant challenges to the generalization ability of existing models. To address this, we propose Cross-Domain Feature Knowledge Distillation (CD-FKD), which enhances the generalization capability of the student network by leveraging both global and instance-wise feature distillation. The proposed method uses diversified data through downscaling and corruption to train the student network, whereas the teacher network receives the original source domain data. The student network mimics the features of the teacher through both global and instance-wise distillation, enabling it to extract object-centric features effectively, even for objects that are difficult to detect owing to corruption. Extensive experiments on challenging scenes demonstrate that CD-FKD outperforms state-of-the-art methods in both target domain generalization and source domain performance, validating its effectiveness in improving object detection robustness to domain shifts. This approach is valuable in real-world applications, like autonomous driving and surveillance, where robust object detection in diverse environments is crucial.

Junseok Lee, Sungho Shin, Seongju Lee +1

Junseok Lee is with the Advanced Robotics Lab, LG Electronics, Seoul, Republic of Korea. · Sungho Shin is with the Place AI Flatform, Naver, Gyeonggi-do, Republic of Korea · Seongju Lee and Kyoobin Lee are with the Department of AI Convergence, Gwangju Institute of Science and Technology (GIST), Gwangju, Republic of Korea.

Object DetectionCross-Domain GeneralizationFeature-Based Knowledge DistillationKnowledge Distillation
AskPDF