Knowledge From the Dark Side: Entropy-Reweighted Knowledge Distillation for Balanced Knowledge Transfer

• URL: http://arxiv.org/abs/2311.13621v1
• Date: Wed, 22 Nov 2023 08:34:33 GMT
• Title: Knowledge From the Dark Side: Entropy-Reweighted Knowledge Distillation for Balanced Knowledge Transfer
• Authors: Chi-Ping Su, Ching-Hsun Tseng, Shin-Jye Lee
• Abstract summary: Distillation (KD) transfers knowledge from a larger "teacher" model to a student. ERKD uses entropy in the teacher's predictions to reweight the KD loss on a sample-wise basis. Our code is available at https://github.com/cpsu00/ER-KD.
• Score: 1.2606200500489302
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Knowledge Distillation (KD) transfers knowledge from a larger "teacher" model to a compact "student" model, guiding the student with the "dark knowledge" $\unicode{x2014}$ the implicit insights present in the teacher's soft predictions. Although existing KDs have shown the potential of transferring knowledge, the gap between the two parties still exists. With a series of investigations, we argue the gap is the result of the student's overconfidence in prediction, signaling an imbalanced focus on pronounced features while overlooking the subtle yet crucial dark knowledge. To overcome this, we introduce the Entropy-Reweighted Knowledge Distillation (ER-KD), a novel approach that leverages the entropy in the teacher's predictions to reweight the KD loss on a sample-wise basis. ER-KD precisely refocuses the student on challenging instances rich in the teacher's nuanced insights while reducing the emphasis on simpler cases, enabling a more balanced knowledge transfer. Consequently, ER-KD not only demonstrates compatibility with various state-of-the-art KD methods but also further enhances their performance at negligible cost. This approach offers a streamlined and effective strategy to refine the knowledge transfer process in KD, setting a new paradigm in the meticulous handling of dark knowledge. Our code is available at https://github.com/cpsu00/ER-KD.

Related papers

• Speculative Knowledge Distillation: Bridging the Teacher-Student Gap Through Interleaved Sampling [81.00825302340984]
  We introduce Speculative Knowledge Distillation (SKD) to generate high-quality training data on-the-fly. In SKD, the student proposes tokens, and the teacher replaces poorly ranked ones based on its own distribution. We evaluate SKD on various text generation tasks, including translation, summarization, math, and instruction following.
  arXiv  Detail & Related papers  (2024-10-15T06:51:25Z)
• Adaptive Explicit Knowledge Transfer for Knowledge Distillation [17.739979156009696]
  We show that the performance of logit-based knowledge distillation can be improved by effectively delivering the probability distribution for the non-target classes from the teacher model. We propose a new loss that enables the student to learn explicit knowledge along with implicit knowledge in an adaptive manner. Experimental results demonstrate that the proposed method, called adaptive explicit knowledge transfer (AEKT) method, achieves improved performance compared to the state-of-the-art KD methods.
  arXiv  Detail & Related papers  (2024-09-03T07:42:59Z)
• Robustness-Reinforced Knowledge Distillation with Correlation Distance and Network Pruning [3.1423836318272773]
  Knowledge distillation (KD) improves the performance of efficient and lightweight models. Most existing KD techniques rely on Kullback-Leibler (KL) divergence. We propose a Robustness-Reinforced Knowledge Distillation (R2KD) that leverages correlation distance and network pruning.
  arXiv  Detail & Related papers  (2023-11-23T11:34:48Z)
• Comparative Knowledge Distillation [102.35425896967791]
  Traditional Knowledge Distillation (KD) assumes readily available access to teacher models for frequent inference. We propose Comparative Knowledge Distillation (CKD), which encourages student models to understand the nuanced differences in a teacher model's interpretations of samples. CKD consistently outperforms state of the art data augmentation and KD techniques.
  arXiv  Detail & Related papers  (2023-11-03T21:55:33Z)
• Towards Understanding and Improving Knowledge Distillation for Neural Machine Translation [59.31690622031927]
  We show that the knowledge comes from the top-1 predictions of teachers. We propose a novel method named textbfTop-1 textbfInformation textbfEnhanced textbfKnowledge textbfDistillation (TIE-KD)
  arXiv  Detail & Related papers  (2023-05-14T08:23:03Z)
• Knowledge Condensation Distillation [38.446333274732126]
  Existing methods focus on excavating the knowledge hints and transferring the whole knowledge to the student. In this paper, we propose Knowledge Condensation Distillation (KCD) Our approach is easy to build on top of the off-the-shelf KD methods, with no extra training parameters and negligible overhead.
  arXiv  Detail & Related papers  (2022-07-12T09:17:34Z)
• Undistillable: Making A Nasty Teacher That CANNOT teach students [84.6111281091602]
  This paper introduces and investigates a concept called Nasty Teacher: a specially trained teacher network that yields nearly the same performance as a normal one. We propose a simple yet effective algorithm to build the nasty teacher, called self-undermining knowledge distillation.
  arXiv  Detail & Related papers  (2021-05-16T08:41:30Z)
• KDExplainer: A Task-oriented Attention Model for Explaining Knowledge Distillation [59.061835562314066]
  We introduce a novel task-oriented attention model, termed as KDExplainer, to shed light on the working mechanism underlying the vanilla KD. We also introduce a portable tool, dubbed as virtual attention module (VAM), that can be seamlessly integrated with various deep neural networks (DNNs) to enhance their performance under KD.
  arXiv  Detail & Related papers  (2021-05-10T08:15:26Z)
• Knowledge Distillation Thrives on Data Augmentation [65.58705111863814]
  Knowledge distillation (KD) is a general deep neural network training framework that uses a teacher model to guide a student model. Many works have explored the rationale for its success, however, its interplay with data augmentation (DA) has not been well recognized so far. In this paper, we are motivated by an interesting observation in classification: KD loss can benefit from extended training iterations while the cross-entropy loss does not. We show this disparity arises because of data augmentation: KD loss can tap into the extra information from different input views brought by DA.
  arXiv  Detail & Related papers  (2020-12-05T00:32:04Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.