EM
Eran Malach
cs.LGstat.MLcs.AIcs.CLcs.NEmath.OCcs.CV
On Valency
published · living versionsW_97xvna8t·v1 · currentpublished
Understanding and Mitigating Role Spoofing in LLMs
with Ezra Edelman, Benjamin Edelman, Surbhi Goel
1 version
W_3r3u6gpt·v1published
Preprints & journals
46 papers in the corpus · 2017–2026Weighted State Souping for Efficient Long-Context Retrieval2610.25621v1 · William L. Tong, Aryo Lotfi, Emmanuel Abbe et al.2026 · 0 citationsarXivon Valency
Fewer Bits, More Tries: Recovering Capability Under Memory Constraints2610.15372v1 · Tomer Galanti, Josh Susskind, Eran Malach2026 · 0 citationsarXiv
On the Importance of Gating: Memorization vs. In-Context Learning in State Space Models2609.16540v1 · William L. Tong, Aryo Lotfi, Emmanuel Abbe et al.2026 · 0 citationsarXiv
Annotations Mitigate Post-Training Mode Collapse2605.09995v1 · Jacob Mitchell Springer, Madhu Advani, Lukas Aichberger et al.2026 · 0 citationsarXiv
LLM Priors for ERM over Programs2510.14331v3 · Shivam Singhal, Priyadarsi Mishra, Eran Malach et al.2025 · 0 citationsarXiv
To Infinity and Beyond: Tool-Use Unlocks Length Generalization in State Space Models2510.14826v1 · Eran Malach, Omid Saremi, Sinead Williamson et al.2025 · 0 citationsarXiv
How Reinforcement Learning After Next-Token Prediction Facilitates Learning2510.11495v2 · Nikolaos Tsilivis, Eran Malach, Karen Ullrich et al.2025 · 0 citationsarXiv
Rethinking JEPA: Compute-Efficient Video SSL with Frozen Teachers2509.24317v1 · Xianhang Li, Chen Huang, Chun-Liang Li et al.2025 · 0 citationsarXiv
A Taxonomy of Transcendence2508.17669v1 · Natalie Abreu, Edwin Zhang, Eran Malach et al.2025 · 0 citationsarXiv
Decomposing Elements of Problem Solving: What "Math" Does RL Teach?2505.22756v1 · Tian Qin, Core Francisco Park, Mujin Kwun et al.2025 · 0 citationsarXiv
Let Me Think! A Long Chain-of-Thought Can Be Worth Exponentially Many Short Ones2505.21825v2 · Parsa Mirtaheri, Ezra Edelman, Samy Jelassi et al.2025 · 0 citationsarXiv
The Power of Random Features and the Limits of Distribution-Free Gradient Descent2505.10423v1 · Ari Karchmer, Eran Malach2025 · 0 citationsarXiv
Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining2504.07912v2 · Rosie Zhao, Alexandru Meterez, Sham Kakade et al.2025 · 0 citationsarXiv
To Backtrack or Not to Backtrack: When Sequential Search Limits Model Reasoning2504.07052v2 · Tian Qin, David Alvarez-Melis, Samy Jelassi et al.2025 · 0 citationsarXiv
The Role of Sparsity for Length Generalization in Transformers2502.16792v1 · Noah Golowich, Samy Jelassi, David Brandfonbrener et al.2025 · 0 citationsarXiv
Loss-to-Loss Prediction: Scaling Laws for All Datasets2411.12925v1 · David Brandfonbrener, Nikhil Anand, Nikhil Vyas et al.2024 · 0 citationsarXiv
Mixture of Parrots: Experts improve memorization more than reasoning2410.19034v2 · Samy Jelassi, Clara Mohri, David Brandfonbrener et al.2024 · 1 citationarXiv
LoRA Soups: Merging LoRAs for Practical Skill Composition Tasks2410.13025v2 · Akshara Prabhakar, Yuanzhi Li, Karthik Narasimhan et al.2024 · 0 citationsarXiv
Don't Stop Me Now: Embedding Based Scheduling for LLMs2410.01035v1 · Rana Shahout, Eran Malach, Chunwei Liu et al.2024 · 0 citationsarXiv
On the Power of Decision Trees in Auto-Regressive Language Modeling2409.19150v1 · Yulu Gan, Tomer Galanti, Tomaso Poggio et al.2024 · 1 citationarXiv
Universal Length Generalization with Turing Programs2407.03310v1 · Kaiying Hou, David Brandfonbrener, Sham Kakade et al.2024 · 0 citationsarXiv
A New Perspective on Shampoo's Preconditioner2406.17748v1 · Depen Morwani, Itai Shapira, Nikhil Vyas et al.2024 · 0 citationsarXiv
Transcendence: Generative Models Can Outperform The Experts That Train Them2406.11741v4 · Edwin Zhang, Vincent Zhu, Naomi Saphra et al.2024 · 6 citationsarXiv
The Evolution of Statistical Induction Heads: In-Context Learning Markov Chains2402.11004v1 · Benjamin L. Edelman, Ezra Edelman, Surbhi Goel et al.2024 · 8 citationsarXiv
Repeat After Me: Transformers are Better than State Space Models at Copying2402.01032v2 · Samy Jelassi, David Brandfonbrener, Sham M. Kakade et al.2024 · 3 citationsarXiv
Auto-Regressive Next-Token Predictors are Universal Learners2309.06979v3 · Eran Malach2023 · 2 citationsarXiv
Pareto Frontiers in Neural Feature Learning: Data, Compute, Width, and Luck2309.03800v2 · Benjamin L. Edelman, Surbhi Goel, Sham Kakade et al.2023 · 0 citationsarXiv
Corgi^2: A Hybrid Offline-Online Approach To Storage-Aware Data Shuffling For SGD2309.01640v1 · Etay Livne, Gal Kaplun, Eran Malach et al.2023 · 0 citationsarXiv
Less is More: Selective Layer Finetuning with SubTuning2302.06354v3 · Gal Kaplun, Andrey Gurevich, Tal Swisa et al.2023 · 1 citationarXiv
Hidden Progress in Deep Learning: SGD Learns Parities Near the Computational Limit2207.08799v3 · Boaz Barak, Benjamin L. Edelman, Surbhi Goel et al.2022 · 15 citationsarXiv
Knowledge Distillation: Bad Models Can Be Good Role Models2203.14649v1 · Gal Kaplun, Eran Malach, Preetum Nakkiran et al.2022 · 7 citationsarXiv
On the Power of Differentiable Learning versus PAC and SQ Learning2108.04190v2 · Emmanuel Abbe, Pritish Kamath, Eran Malach et al.2021 · 1 citationarXiv
Quantifying the Benefit of Using Differentiable Learning over Tangent Kernels2103.01210v1 · Eran Malach, Pritish Kamath, Emmanuel Abbe et al.2021 · 4 citationsarXiv
The Connection Between Approximation, Depth Separation and Learnability in Neural Networks2102.00434v2 · Eran Malach, Gilad Yehudai, Shai Shalev-Shwartz et al.2021 · 5 citationsarXiv
Computational Separation Between Convolutional and Fully-Connected Networks2010.01369v1 · Eran Malach, Shai Shalev-Shwartz2020 · 16 citationsarXiv
When Hardness of Approximation Meets Hardness of Learning2008.08059v2 · Eran Malach, Shai Shalev-Shwartz2020 · 2 citationsarXiv
Learning Parities with Neural Networks2002.07400v2 · Amit Daniely, Eran Malach2020 · 17 citationsarXiv
Proving the Lottery Ticket Hypothesis: Pruning is All You Need2002.00585v1 · Eran Malach, Gilad Yehudai, Shai Shalev-Shwartz et al.2020 · 127 citationsarXiv
Learning Boolean Circuits with Neural Networks1910.11923v2 · Eran Malach, Shai Shalev-Shwartz2019 · 2 citationsarXiv
On the Optimality of Trees Generated by ID31907.05444v2 · Alon Brutzkus, Amit Daniely, Eran Malach2019 · 6 citationsarXiv
ID3 Learns Juntas for Smoothed Product Distributions1906.08654v1 · Alon Brutzkus, Amit Daniely, Eran Malach2019 · 10 citationsarXiv
Decoupling Gating from Linearity1906.05032v1 · Jonathan Fiat, Eran Malach, Shai Shalev-Shwartz2019 · 10 citationsarXiv
Is Deeper Better only when Shallow is Good?1903.03488v1 · Eran Malach, Shai Shalev-Shwartz2019 · 22 citationsarXiv
A Provably Correct Algorithm for Deep Learning that Actually Works1803.09522v2 · Eran Malach, Shai Shalev-Shwartz2018 · 17 citationsarXiv
SGD Learns Over-parameterized Networks that Provably Generalize on Linearly Separable Data1710.10174v1 · Alon Brutzkus, Amir Globerson, Eran Malach et al.2017 · 68 citationsarXiv
Decoupling "when to update" from "how to update"1706.02613v2 · Eran Malach, Shai Shalev-Shwartz2017 · 368 citationsarXiv
Career total: 50 works. 46 are in this corpus.
Profile built from the corpus for this byline.
Author records are still filling in while the Hub is in alpha. If this is your page, you'll be able to claim it soon. Spot a mistake? Tell us.