paper-with-me

홈 › Papers

Dynamic Temperature Scheduler for Knowledge Distillation

2025-11-14 · Sibgat Ul Islam, Jawad Ibn Ahad, Fuad Rahman, Mohammad Ruhul Amin, Nabeel Mohammed, Shafin Rahman arxiv

Knowledge Distillation (KD) trains a smaller student model using a large, pre-trained teacher model, with temperature as a key hyperparameter controlling the softness of output probabilities. Traditional methods use a fixed temperature throughout training, which is suboptimal. Moreover, architectural differences between teacher and student often result in mismatched logit magnitudes. We demonstrate that students benefit from softer probabilities early in training but require sharper probabilities in later stages. We introduce Dynamic Temperature Scheduler (DTS), which adjusts temperature dynamically based on the cross-entropy loss gap between teacher and student. To our knowledge, this is the first temperature scheduling method that adapts based on the divergence between teacher and student distributions. Our method integrates seamlessly with existing KD frameworks. We validate DTS across multiple KD strategies on vision (CIFAR-100, Tiny-ImageNet) and NLP tasks (GLUE, Dolly, SelfIns, UnNI, S-NI), consistently outperforming static-temperature baselines. Code is available at https://github.com/Sibgat-Ul/DTS.

📄 PDF Abstract BibTeX arXiv:2511.13767

Code (0)

등록된 구현이 없습니다.

Tasks

Knowledge Distillation

Similar Papers 제목 키워드 기반

HPM-KD: Hierarchical Progressive Multi-Teacher Framework for Knowledge Distillation and Efficient Model Compression

2025-12-10 · Gustavo Coelho Haase, Paulo Henrique Dourado da Silva arxiv

Knowledge Distillation (KD) has emerged as a promising technique for model compression but faces critical limitations: (1) sensitivity to hyperparameters requiring extensive manual tuning, (2) capacity gap when distillin…

Knowledge DistillationModel Compression

Dynamic Temperature Knowledge Distillation

2024-04-19 · Yukang Wei, Yu Bai

Temperature plays a pivotal role in moderating label softness in the realm of knowledge distillation (KD). Traditional approaches often employ a static temperature throughout the KD process, which fails to address the nu…

Knowledge Distillation

Curriculum Temperature for Knowledge Distillation

2022-11-29 · Zheng Li, Xiang Li, Lingfeng Yang, Borui Zhao 외

Most existing distillation methods ignore the flexible role of the temperature in the loss function and fix it as a hyper-parameter that can be decided by an inefficient grid search. In general, the temperature controls …

Image ClassificationKnowledge Distillation

LLM-Oriented Token-Adaptive Knowledge Distillation

2025-10-13 · Xurong Xie, Zhucun Xue, Jiafu Wu, Jian Li 외 arxiv

Knowledge distillation (KD) is a key technique for compressing large-scale language models (LLMs), yet prevailing logit-based methods typically employ static strategies that are misaligned with the dynamic learning proce…

Knowledge Distillation

AdaAnn: Adaptive Annealing Scheduler for Probability Density Approximation

2022-02-01 · Emma R. Cobian, Jonathan D. Hauenstein, Fang Liu, Daniele E. Schiavazzi

Approximating probability distributions can be a challenging task, particularly when they are supported over regions of high geometrical complexity or exhibit multiple modes. Annealing can be used to facilitate this task…

Computational Efficiencyparameter estimationVariational Inference