Noisy expert imitation learning requires exponential samples for offline methods but polynomial for a variant of on-policy distillation under a noise condition.
super hub Canonical reference
Distilling the Knowledge in a Neural Network
Canonical reference. 79% of citing Pith papers cite this work as background.
680
Pith papers citing it
Background
79% of classified citations
abstract
A very simple way to improve the performance of almost any machine learning algorithm is to train many different models on the same data and then to average their predictions. Unfortunately, making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to a large number of users, especially if the individual models are large neural nets. Caruana and his collaborators have shown that it is possible to compress the knowledge in an ensemble into a single model which is much easier to deploy and we develop this approach further using a different compression technique. We achieve some surprising results on MNIST and we show that we can significantly improve the acoustic model of a heavily used commercial system by distilling the knowledge in an ensemble of models into a single model. We also introduce a new type of ensemble composed of one or more full models and many specialist models which learn to distinguish fine-grained classes that the full models confuse. Unlike a mixture of experts, these specialist models can be trained rapidly and in parallel.
hub tools
citation-role summary
background 70
method 14
other 2
dataset 1
citation-polarity summary
claims ledger
- abstract A very simple way to improve the performance of almost any machine learning algorithm is to train many different models on the same data and then to average their predictions. Unfortunately, making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to a large number of users, especially if the individual models are large neural nets. Caruana and his collaborators have shown that it is possible to compress the knowledge in an ensemble into a single model which is much easier to deploy and we develop this approach further using
authors
co-cited works
representative citing papers
A formal game-based study establishes that black-box proofs of ownership for ML classifiers are possible precisely when the concept class is not self-correctable.
PrimeKG-CL supplies the first continual graph learning benchmark using authentic temporal snapshots from nine biomedical databases, showing strong interactions between embedding decoders and learning strategies plus limits of standard metrics on retention versus forgetting.
Inference-time refinement of pre-trained tabular diffusion models via Bidirectional Chamfer Refinement achieves median 8.6% better downstream performance than real data across 15 benchmarks while preserving fidelity and privacy.
Cayley unitary adapters executed on real quantum hardware improve LLM perplexity by 1.4% on Llama 3.1 8B with 6000 parameters and recover 83% of compression-induced degradation on SmolLM2.
Tiny language models under 10M parameters trained on a synthetic children's story dataset generate fluent, consistent, multi-paragraph English text with near-perfect grammar and reasoning.
Self-supervised ViTs show emergent semantic segmentation and 78.3% k-NN accuracy on ImageNet; DINO reaches 80.1% linear evaluation with ViT-Base.
GPT-3 shows that scaling an autoregressive language model to 175 billion parameters enables strong few-shot performance across diverse NLP tasks via in-context prompting without fine-tuning.
Purified OPSD subtracts a reference-only teacher's signal from standard OPSD supervision and applies PMI to create a cleaner distillation target, yielding gains on long-CoT models while preserving epistemic behavior.
Empirical study on production-scale clinical NLP shows direct learning from verifier rejections fails due to sparse data while fixed ontology and evidence-support filters succeed, with selectivity determined by matching verifier evidence.
TallyTrain is a hard-label distillation protocol for federated learning that uses argmax transmission and optional sparse merges to match soft-label performance at up to 1000x lower communication cost.
Cortex uses an Ontological Corpus Graph to structure web-scale corpora, creating a refined 24.14B-token corpus and a new benchmark validated on eight LLMs.
RPM-Distill uses synchronized radar only at training time to distill spectral periodic features into a video model via adaptive per-sample gating, yielding 81% lower MAE on remote physiological measurement tasks.
BiLoc is the first binary neural network framework for 6-DoF LiDAR pose estimation that uses an auxiliary objective to adaptively regulate information retention and achieve SOTA among BNNs on large outdoor datasets.
LaViD distills LLM conceptual knowledge to vision models via LLM-generated MCQ soft labels, outperforming vision-language distillation baselines on fine-grained benchmarks while improving robustness on spurious correlation datasets.
REDI-Match uses rotation-equivariant distillation to transfer VFM semantics into a strictly equivariant encoder plus an entropy-driven alignment module, claiming SOTA accuracy and 1.9x speed on rotation-heavy benchmarks.
SC-GRPO improves RL with verifiable rewards by multiplying GRPO gradients with self-induced per-token KL divergence, outperforming GRPO by 8.1% and DAPO by 5.9% on math, code, and agent benchmarks.
A calibration strategy using full-Jones corrections with an in-field unpolarised calibrator and visibility-based multi-epoch alignment enables sub-arcsecond polarimetric imaging with LOFAR at metre wavelengths.
ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite with largest gains at the 0.8B scale.
d-OPSD reframes on-policy self-distillation for dLLMs via suffix conditioning from self-generated answers and step-level supervision, outperforming RLVR and SFT on reasoning benchmarks with ~10% of the optimization steps.
Linear recoverability of transformer FFN blocks varies widely across depth, is learned during training, and is independent of the activation function.
Doc-to-Atom decomposes documents into composable micro-LoRA adapters selected by a query router for efficient long-context QA.
Self-distillation from a caption-conditioned video diffusion model to an image-and-prompt-conditioned executor, enhanced by RL from VLM feedback, enables task solving in world models.
A stop-gradient consistency regularizer mitigates context-induced degradation in on-policy distillation, improving robustness across 12 configurations.
citing papers explorer
-
Behavior Cloning is Not All You Need: The Optimality of On-Policy Distillation for Noisy Expert Feedback
Noisy expert imitation learning requires exponential samples for offline methods but polynomial for a variant of on-policy distillation under a noise condition.
-
Proofs of Ownership for Machine Learning Models
A formal game-based study establishes that black-box proofs of ownership for ML classifiers are possible precisely when the concept class is not self-correctable.
-
PrimeKG-CL: A Continual Graph Learning Benchmark on Evolving Biomedical Knowledge Graphs
PrimeKG-CL supplies the first continual graph learning benchmark using authentic temporal snapshots from nine biomedical databases, showing strong interactions between embedding decoders and learning strategies plus limits of standard metrics on retention versus forgetting.
-
Inference-Time Refinement Closes the Synthetic-Real Gap in Tabular Diffusion
Inference-time refinement of pre-trained tabular diffusion models via Bidirectional Chamfer Refinement achieves median 8.6% better downstream performance than real data across 15 benchmarks while preserving fidelity and privacy.
-
Quantum-enhanced Large Language Models on Quantum Hardware via Cayley Unitary Adapters
Cayley unitary adapters executed on real quantum hardware improve LLM perplexity by 1.4% on Llama 3.1 8B with 6000 parameters and recover 83% of compression-induced degradation on SmolLM2.
-
TinyStories: How Small Can Language Models Be and Still Speak Coherent English?
Tiny language models under 10M parameters trained on a synthetic children's story dataset generate fluent, consistent, multi-paragraph English text with near-perfect grammar and reasoning.
-
Emerging Properties in Self-Supervised Vision Transformers
Self-supervised ViTs show emergent semantic segmentation and 78.3% k-NN accuracy on ImageNet; DINO reaches 80.1% linear evaluation with ViT-Base.
-
Language Models are Few-Shot Learners
GPT-3 shows that scaling an autoregressive language model to 175 billion parameters enables strong few-shot performance across diverse NLP tasks via in-context prompting without fine-tuning.
-
Purified OPSD: On-Policy Self-Distillation Without Losing How to Think
Purified OPSD subtracts a reference-only teacher's signal from standard OPSD supervision and applies PMI to create a cleaner distillation target, yielding gains on long-CoT models while preserving epistemic behavior.
-
Dynamic Bidirectional Pattern Memory: A Production-Scale Empirical Characterisation of Inference-Time Gating in Clinical NLP
Empirical study on production-scale clinical NLP shows direct learning from verifier rejections fails due to sparse data while fixed ontology and evidence-support filters succeed, with selectivity determined by matching verifier evidence.
-
TallyTrain: Communication-Efficient Federated Distillation
TallyTrain is a hard-label distillation protocol for federated learning that uses argmax transmission and optional sparse merges to match soft-label performance at up to 1000x lower communication cost.
-
CORTEX: High-Quality Cross-Domain Organization of Web-Scale Corpora through Ontological Corpus Graph
Cortex uses an Ontological Corpus Graph to structure web-scale corpora, creating a refined 24.14B-token corpus and a new benchmark validated on eight LLMs.
-
RPM-Distill: Physiology-guided Adaptive Cross-modal Distillation for Robust Remote Physiological Measurement
RPM-Distill uses synchronized radar only at training time to distill spectral periodic features into a video model via adaptive per-sample gating, yielding 81% lower MAE on remote physiological measurement tasks.
-
Learning 1-Bit LiDAR-based Localization with Auxiliary Objective
BiLoc is the first binary neural network framework for 6-DoF LiDAR pose estimation that uses an auxiliary objective to adaptively regulate information retention and achieve SOTA among BNNs on large outdoor datasets.
-
Large Language Model Teaches Visual Students: Cross-Modality Transfer of Fine-Grained Conceptual Knowledge
LaViD distills LLM conceptual knowledge to vision models via LLM-generated MCQ soft labels, outperforming vision-language distillation baselines on fine-grained benchmarks while improving robustness on spurious correlation datasets.
-
REDI-Match: Rotation-Equivariant Distillation for Efficient and Robust Dense Matching
REDI-Match uses rotation-equivariant distillation to transfer VFM semantics into a strictly equivariant encoder plus an entropy-driven alignment module, claiming SOTA accuracy and 1.9x speed on rotation-heavy benchmarks.
-
Learning from Own Solutions: Self-Conditioned Credit Assignment for Reinforcement Learning with Verifiable Rewards
SC-GRPO improves RL with verifiable rewards by multiplying GRPO gradients with self-induced per-token KL divergence, outperforming GRPO by 8.1% and DAPO by 5.9% on math, code, and agent benchmarks.
-
Polarisation and Faraday rotation measure imaging at metre wavelengths with sub-arcsecond resolution: a foundational calibration strategy
A calibration strategy using full-Jones corrections with an in-field unpolarised calibrator and visibility-based multi-epoch alignment enables sub-arcsecond polarimetric imaging with LOFAR at metre wavelengths.
-
Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients
ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite with largest gains at the 0.8B scale.
-
Learning from the Self-future: On-policy Self-distillation for dLLMs
d-OPSD reframes on-policy self-distillation for dLLMs via suffix conditioning from self-generated answers and step-level supervision, outperforming RLVR and SFT on reasoning benchmarks with ~10% of the optimization steps.
-
How Linear Is a Transformer Feed-Forward Block? Per-Block Linear Recoverability Is Learned, Not Architectural
Linear recoverability of transformer FFN blocks varies widely across depth, is learned during training, and is independent of the activation function.
-
Doc-to-Atom: Learning to Compile and Compose Memory Atoms
Doc-to-Atom decomposes documents into composable micro-LoRA adapters selected by a query router for efficient long-context QA.
-
World Model Self-Distillation: Training World Models to Solve General Tasks
Self-distillation from a caption-conditioned video diffusion model to an image-and-prompt-conditioned executor, enhanced by RL from VLM feedback, enables task solving in world models.
-
When Context Returns: Toward Robust Internalization in On-Policy Distillation
A stop-gradient consistency regularizer mitigates context-induced degradation in on-policy distillation, improving robustness across 12 configurations.
-
Quo Vadis, Visual In-Context Learning? A Unified Benchmark Across Domains and Tasks
The paper constructs the VIBE benchmark and evaluates six visual in-context learning models on 14 datasets, 12 tasks, and 106 combinations under a unified one-shot protocol, revealing limitations and failure modes.
-
Beyond Absolute Imitation: Anchored Residual Guidance for Privileged On-Policy Distillation
AR-OPD disentangles privileged supervision via anchored residual guidance to reduce hindsight leakage in on-policy distillation, reporting gains of 2.3 points over full privileged OPD and 7.9 over SFT on reasoning tasks.
-
Quantifying and Defending against the Privacy Risk in Logit-based Federated Learning
Logit-based federated learning leaks private model information to a semi-honest server via shared logits even with unrelated public data, enabling an adaptive stealing attack with theoretical bounds and a logit-perturbation defense.
-
Fast LLM-Based Semantic Filtering: From a Unified Framework to an Adaptive Two-Phase Method
An adaptive two-phase semantic filter using clustering then a hybrid proxy trained on LLM confidence achieves 1.6-2.0x speedup over prior methods at 90% accuracy on 10K document corpora.
-
Towards Tight Bounds for Streaming Attention
The paper closes the gap between upper and lower bounds on space for streaming attention approximation by combining discrepancy, polynomial, and partitioning techniques for algorithms and a new INDEX-based lower bound method.
-
OPRD: On-Policy Representation Distillation
OPRD performs distillation in hidden-state space on on-policy data for deterministic gradients and better math benchmark performance, plus OPRD-Bridge for cross-architecture transfer via low-rank projectors.
-
Reinforcement Learning from Rich Feedback with Distributional DAgger
DistIL applies distributional DAgger with forward cross-entropy to achieve monotonic policy improvement and better Pass@N from rich feedback in RL for reasoning tasks.
-
Toward Calibrated, Fair, and accurate Deepfake Detection
Face-Feature Tuning is a label-free logit remapping method that reduces FPR/TPR gaps across groups in deepfake detection while preserving overall accuracy.
-
RogueMerge: Robust and Unified Attacks against LLM Model Merging
RogueMerge is a unified attack method that jointly optimizes task vectors to succeed after merging, using stochastic min-max simulation for unknown merging settings and a Taylor-approximated DRO for prompt generalization on generative LLMs.
-
TAPS: Target-Aware Prefix Tree Selection for Diffusion-Drafted Speculative Decoding
TAPS converts diffusion marginal probabilities into path-conditioned acceptance estimates to select prefix-closed subtrees under a fixed verification budget, achieving up to 7.9x end-to-end speedup over autoregressive decoding.
-
Ego-METAS: Egocentric online Multimodal Energy-efficient Temporal Action Segmentation benchmark
Ego-METAS is a new benchmark providing unified egocentric video data, splits, features and baselines for online multimodal temporal action segmentation under hardware-representative energy constraints.
-
Bounded Behavioral Indistinguishability for Black-Box LLM Distillation
Introduces (ε,q,t,A)-behavioral indistinguishability and shows via Qwen/Llama experiments that LoRA distillation boosts semantic similarity but leaves detectable behavioral differences under adversarial evaluation.
-
Quantum Subliminal Learning
QNNs retain most hidden-task signals through public-task interfaces while classical networks transmit little, with transmission governed by teacher drift magnitude and the visible fraction of hidden drift in a unified geometric model.
-
FedQHD: Closed-Form Function-Space Federated Reinforcement Learning
FedQHD achieves closed-form federated Q-learning via hyperdimensional encoders with linear readouts, formalizes the federation gap under heterogeneous encoders, and reports competitive performance on continuous-state benchmarks with reduced computation.
-
Beyond Recall: Behavioral Specification as an Interpretive Layer for AI Personalization
A Behavioral Specification interpretive layer improves representational accuracy for AI personalization by compressing user data into patterns, outperforming raw corpora and commercial memory systems on held-out behavioral predictions across 14 autobiographical corpora while reducing context cost.
-
Auditing and Fixing Economic Validity in Tabular Foundation Models for Discrete Choice
A two-stage adapter embeds tabular foundation model predictions in a utility-maximization framework to achieve up to 13 percentage points higher accuracy than logit models while guaranteeing economic consistency on transportation datasets.
-
MuCRASP: Multimodal Chain-of-thought Reasoning aware Structured Pruning
MuCRASP prunes VLMs in a CoT-aware manner, outperforming baselines by preserving reasoning quality at 30-50% compression rates on models like Qwen2.5-VL-7B.
-
Learning Through Noise: Why Subliminal Learning Works and When It Fails
Subliminal learning occurs via compatible auxiliary and class output heads on task-unrelated inputs, even with random hidden layers or architecture changes, with theory and upper bounds on failure.
-
Slimmable ConvNeXt: Width-Adaptive Inference for Efficient Multi-Device Deployment
Slimmable ConvNeXt adapts ConvNeXt for width-adaptive inference using LayerNorm and inverted bottlenecks, reaching 80.8% top-1 at 4.5 GMACs and outperforming HydraViT, MatFormer, and SortedNet on ImageNet-1k.
-
Visual-Advantage On-Policy Distillation for Vision-Language Models
VA-OPD improves VLM performance over standard on-policy distillation by reweighting rollouts and separating KL terms according to token-level visual advantage on math and visual benchmarks.
-
X-Token: Projection-Guided Cross-Tokenizer Knowledge Distillation
X-Token proposes projection-guided P-KL and H-KL losses to fix uncommon-token suppression and over-conservative matching in logit-based cross-tokenizer distillation, yielding gains over GOLD on Llama-3.2-1B.
-
Layer-wise Token Compression for Efficient Document Reranking
Layer-wise Token Compression applies adaptive token pooling at middle transformer layers for cross-encoder rerankers, preserving MS MARCO ranking quality while raising QPS up to 25% on passages and 116% on documents, with added gains on listwise LLM rerankers and a regularizer effect for long inputs
-
Faster or Stronger: Towards Flexible Visual Place Recognition via Weighted Aggregation and Token Pruning
Proposes weighted aggregation of clusters and self-distillation-driven token pruning to improve both accuracy and efficiency in ViT-based visual place recognition.
-
Code Generation by Differential Test Time Scaling
DiffCodeGen clusters code candidates by behavioral similarity from fuzzing-synthesized inputs and selects the largest cluster's medoid, matching or exceeding prior test-time scaling methods with far less token and time cost.
-
When Does Model Collapse Occur in Structured Interactive Learning?
Model collapse occurs in structured interactive learning if and only if the directed interaction graph satisfies a specific topological condition, with finite-sample guarantees for linear regression and asymptotic results for M-estimators.
-
Draft Less, Retrieve More: Hybrid Tree Construction for Speculative Decoding
Graft combines pruning and retrieval in a sequential mechanism to build hybrid draft trees for speculative decoding, delivering up to 5.41× speedup and 21.8% better average speedup than EAGLE-3 on large models.