Follow
Kaifeng Lyu
Kaifeng Lyu
Verified email at princeton.edu - Homepage
Title
Cited by
Cited by
Year
Gradient Descent Maximizes the Margin of Homogeneous Neural Networks
K Lyu, J Li
2020 International Conference on Learning Representations (ICLR 2020), 2020
2832020
Theoretical analysis of auto rate-tuning by batch normalization
S Arora, Z Li, K Lyu
2019 International Conference on Learning Representations (ICLR 2019), 2019
1162019
Learning gradient descent: Better generalization and longer horizons
K Lv, S Jiang, J Li
34th International Conference on Machine Learning (ICML 2017) 70, 2247-2255, 2017
1082017
Towards Resolving the Implicit Bias of Gradient Descent for Matrix Factorization: Greedy Low-Rank Learning
Z Li, Y Luo, K Lyu
2021 International Conference on Learning Representations (ICLR 2021), 2021
1072021
Gradient Descent on Two-layer Nets: Margin Maximization and Simplicity Bias
K Lyu, Z Li, R Wang, S Arora
35th Conference on Neural Information Processing Systems (NeurIPS 2021), 2021
582021
Reconciling Modern Deep Learning with Traditional Optimization Analyses: The Intrinsic Learning Rate
Z Li, K Lyu, S Arora
34th Conference on Neural Information Processing Systems (NeurIPS 2020), 2020
582020
Understanding the generalization benefit of normalization layers: Sharpness reduction
K Lyu, Z Li, S Arora
36th Conference on Neural Information Processing Systems (NeurIPS 2022), 2022
522022
Fine-grained complexity meets IP = PSPACE
L Chen, S Goldwasser, K Lyu, GN Rothblum, A Rubinstein
30th Annual ACM-SIAM Symposium on Discrete Algorithms (SODA 2019), 1-20, 2019
372019
Understanding incremental learning of gradient descent: A fine-grained analysis of matrix sensing
J Jin, Z Li, K Lyu, SS Du, JD Lee
International Conference on Machine Learning (ICML 2023), 15200-15238, 2023
232023
On the SDEs and Scaling Rules for Adaptive Gradient Algorithms
S Malladi, K Lyu, A Panigrahi, S Arora
36th Conference on Neural Information Processing Systems (NeurIPS 2022), 2022
222022
DistillSpec: Improving speculative decoding via knowledge distillation
Y Zhou, K Lyu, AS Rawat, AK Menon, A Rostamizadeh, S Kumar, JF Kagy, ...
2024 International Conference on Learning Representations (ICLR 2024), 2023
162023
Why (and When) does Local SGD Generalize Better than SGD?
X Gu, K Lyu, L Huang, S Arora
2023 International Conference on Learning Representations (ICLR 2023), 2023
132023
Dichotomy of Early and Late Phase Implicit Biases Can Provably Induce Grokking
K Lyu, J Jin, Z Li, SS Du, JD Lee, W Hu
2024 International Conference on Learning Representations (ICLR 2024), 2023
62023
Single-Source Bottleneck Path Algorithm Faster than Sorting for Sparse Graphs
R Duan, K Lyu, H Wu, Y Xie
45th International Colloquium on Automata, Languages, and Programming (ICALP …, 2018
62018
New Definitions and Evaluations for Saliency Methods: Staying Intrinsic, Complete and Sound
A Gupta, N Saunshi, D Yu, K Lyu, S Arora
36th Conference on Neural Information Processing Systems (NeurIPS 2022), 2022
42022
The marginal value of momentum for small learning rate SGD
R Wang, S Malladi, T Wang, K Lyu, Z Li
2024 International Conference on Learning Representations (ICLR 2024), 2023
32023
RNNs are not Transformers (Yet): The Key Bottleneck on In-context Retrieval
K Wen, X Dang, K Lyu
arXiv preprint arXiv:2402.18510, 2024
2024
Keeping LLMs Aligned After Fine-tuning: The Crucial Role of Prompt Templates
K Lyu, H Zhao, X Gu, D Yu, A Goyal, S Arora
arXiv preprint arXiv:2402.18540, 2024
2024
Efficient Stagewise Pretraining via Progressive Subnetworks
A Panigrahi, N Saunshi, K Lyu, S Miryoosefi, S Reddi, S Kale, S Kumar
arXiv preprint arXiv:2402.05913, 2024
2024
A Quadratic Synchronization Rule for Distributed Deep Learning
X Gu, K Lyu, S Arora, J Zhang, L Huang
2024 International Conference on Learning Representations (ICLR 2024), 2023
2023
The system can't perform the operation now. Try again later.
Articles 1–20