MLSP-P64.9
A DUAL-PATH APPROACH TO OPTIMIZING LLMS: ENTROPY CONSTRAINT FOR EXPLOITATION AND NEURAL PERTURBATION FOR EXPLORATION
Dongxiang Zhang, South China University of Technology, China; Yiming Li, Pengcheng Laboratory, China; Erkun Zhang, Fangjiong Chen, South China University of Technology, China; Yijia Zhang, Shixun Zhang, Pengcheng Laboratory, China
Session:
MLSP-P64: Efficient Training and Adaptation of Large Language Models II Poster
Track:
Machine Learning for Signal Processing [ML]
Location:
Poster Area 6
Presentation Time:
Thu, 7 May, 16:30 - 18:30
Presentation
Discussion
Resources
No resources available.
Session MLSP-P64
MLSP-P64.1: FIBKD: A FIBER BUNDLE-BASED FRAMEWORK FOR EFFECTIVE KNOWLEDGE DISTILLATION
Junjie Wu, Fanzhang Li, Soochow University, China
MLSP-P64.2: BORA: BLOCKWISE ORTHOGONAL RANK-1 ADAPTIVE OPTIMIZATION
Yihan Zhang, Jialiang Wang, State Key Laboratory of Advanced Vehicle Integration and Control, China FAW, China
MLSP-P64.3: SCALING SENTIMENT STRENGTH VIA SENTIMENT MIXING
Zhongquan Jian, Minjiang University, China; Ke Yao, Bingbing Hu, Shaopan Wang, Qingqiang Wu, Junfeng Yao, Xiamen University, China
MLSP-P64.4: SAD-SAM: MULTIDIMENSIONAL DISTRIBUTION-ALIGNED SPATIAL-AWARE DISTILLATION FOR SEGMENT ANYTHING MODEL
Yixing Ma, Chengliang Wang, Xing Wu, Chongqing University, China; Hongqian Wang, Peng Wang, The First Affiliated Hospital of Army Medical University, China; Hao Wu, The First Affiliated Hospital of Chongqing Medical University, China
MLSP-P64.5: HYBRID ZEROTH-ORDER FINE-TUNING FOR LANGUAGE MODEL WITH CPU MEMORY ASSISTANCE
Dingkai Wang, Yunnan University, China
MLSP-P64.6: PARAMETER-FREE MIXTURE OF EXPERTS FOR BLACK-BOX PROMPT TUNING
Xiaolin Dong, Lizheng Liu, Shuai Gong, Chaoran Cui, Shandong University of Finance and Economics, China
MLSP-P64.7: PLPP: PROMPT LEARNING WITH PERPLEXITY IS SELF-DISTILLATION FOR VISION-LANGUAGE MODELS
Biao Liu, Southern University of Science and Technology, China; Wenyi Fang, Xiaoyu Wu, Yang Zheng, Zheng Hu, Huawei, China; Bo Yuan, Southern University of Science and Technology, China
MLSP-P64.8: NOT ALL WEIGHT VECTORS ARE NEEDED: COVARIANCE-BASED VECTOR SELECTION TUNING FOR LARGE LANGUAGE MODELS
Lanyu Zheng, Chun Guan, Zhongxue Gan, Siyang Leng, Fudan University, China
MLSP-P64.9: A DUAL-PATH APPROACH TO OPTIMIZING LLMS: ENTROPY CONSTRAINT FOR EXPLOITATION AND NEURAL PERTURBATION FOR EXPLORATION
Dongxiang Zhang, South China University of Technology, China; Yiming Li, Pengcheng Laboratory, China; Erkun Zhang, Fangjiong Chen, South China University of Technology, China; Yijia Zhang, Shixun Zhang, Pengcheng Laboratory, China
MLSP-P64.10: RESOLVING LOW-RANK UPDATE LIMITATIONS FOR MEMORY-EFFICIENT VISUAL NEURAL NETWORK TRAINING
Yuchuan Wang, Jingyu Wang, Qi Qi, Zirui Zhuang, Haifeng Sun, Bo He, Yuhao Li, jianxin Liao, Beijing University of Posts and Telecommunications, China
Contacts