14 papers
Narrowing the Collaboration Gap, Probably
Mirah Shi, Marcel Hussing, Ira Globus-Harris +3
Large language models are increasingly deployed as teams of agents that hold different private information about a shared task. How should such agents concisely communicate in a wa…
LESSER: Post-Training Data Selection with Output-Layer Gradients
Lyuxin David Zhang, Eric Wong, Surbhi Goel +1
The choice of post-training data for large language models substantially affects downstream performance. Gradient-based data selection is a popular approach that ranks training dat…
Model Agreement via Anchoring
Eric Eaton, Surbhi Goel, Marcel Hussing +4
Numerous lines of aim to control -- the extent to which two machine learning models disagree in their predictions. We adopt a simple and standard noti…
Personalization Aids Pluralistic Alignment Under Competition
Natalie Collina, Surbhi Goel, Aaron Roth +1
Can competition among misaligned AI providers yield aligned outcomes for a diverse population of users, and what role does model personalization play? We study a setting where mult…
Why Do Transformers Fail to Forecast Time Series In-Context?
Yufa Zhou, Yixiao Wang, Surbhi Goel +1
Time series forecasting (TSF) remains a challenging and largely unsolved problem in machine learning, despite significant recent efforts leveraging Large Language Models (LLMs), wh…
Emergent Alignment via Competition
Natalie Collina, Surbhi Goel, Aaron Roth +2
Aligning AI systems with human values remains a fundamental challenge, but does our inability to create perfectly aligned models preclude obtaining the benefits of alignment? We st…
Conformal Language Model Reasoning with Coherent Factuality
Maxon Rubin-Toles, Maya Gambhir, Keshav Ramji +2
Language models are increasingly being used in important decision pipelines, so ensuring the correctness of their outputs is crucial. Recent work has proposed evaluating the "factu…
Collaborative Prediction: Tractable Information Aggregation via Agreement
Natalie Collina, Ira Globus-Harris, Surbhi Goel +3
We give efficient "collaboration protocols" through which two parties, who observe different features about the same instances, can interact to arrive at predictions that are more…
Probabilistic Stability Guarantees for Feature Attributions
Helen Jin, Anton Xue, Weiqiu You +2
Stability guarantees have emerged as a principled way to evaluate feature attributions, but existing certification methods rely on heavily smoothed classifiers and often produce co…
Tractable Agreement Protocols
Natalie Collina, Surbhi Goel, Varun Gupta +1
We present an efficient reduction that converts any machine learning algorithm into an interactive protocol, enabling collaboration with another party (e.g., a human) to achieve co…
Logicbreaks: A Framework for Understanding Subversion of Rule-based Inference
Anton Xue, Avishree Khare, Rajeev Alur +2
We study how to subvert large language models (LLMs) from following prompt-specified rules. We first formalize rule-following as inference in propositional Horn logic, a mathematic…
Explicitly Encoding Structural Symmetry is Key to Length Generalization in Arithmetic Tasks
Mahdi Sabbaghi, George Pappas, Hamed Hassani +1
Despite the success of Transformers on language understanding, code generation, and logical reasoning, they still fail to generalize over length on basic arithmetic tasks such as a…
Stochastic Bandits with ReLU Neural Networks
Kan Xu, Hamsa Bastani, Surbhi Goel +1
We study the stochastic bandit problem with ReLU neural network structure. We show that a regret guarantee is achievable by considering bandits with one-layer…
Complexity Matters: Dynamics of Feature Learning in the Presence of Spurious Correlations
GuanWen Qiu, Da Kuang, Surbhi Goel
Existing research often posits spurious features as easier to learn than core features in neural network optimization, but the impact of their relative simplicity remains under-exp…