ARORA Research Lab
Our research lab develops conceptual understanding of AI models, including training techniques, datasets, model interpretability and model evaluations. Recent works involve new ways of training and evaluating models using "skills", which are themselves elicited from powerful AI models, and in turn can be used to generate very effective pipelines for improving AI models using synthetic training data. Many of our papers involve mathematical analysis as well as experiments.
News
Archive →
05/2025
Paper
New paper On the Power of Context-Enhanced Learning in LLMs was accepted to ICML 2025 (Spotlight).
05/2025
Paper
New paper Generalizing from SIMPLE to HARD Visual Reasoning: Can We Mitigate Modality Imbalance in VLMs? was accepted to ICML 2025.
05/2025
Paper
Check out our new paper Why is Your Language Model a Poor Implicit Reward Model?.
03/2025
Award
Congrats to Abhishek Panigrahi on being named 2025 Apple Scholar in AIML!
03/2025
Paper
Check out our new paper What Makes a Reward Model a Good Teacher? An Optimization Perspective.
02/2025
Paper
Check out our new paper Goedel-Prover: A Frontier Model for Open-Source Automated Theorem Proving.
02/2025
Paper
Check out our new paper Unrealized Expectations: Comparing AI Methods vs Classical Algorithms for Maximum Independent Set.
01/2025
Paper
New paper Instruct-SkillMix: A Powerful Pipeline for LLM Instruction Tuning was accepted to ICLR 2025.
01/2025
Paper
New paper Progressive distillation induces an implicit curriculum was accepted to ICLR 2025 (Oral).
01/2025
Paper
New paper Unintentional Unalignment: Likelihood Displacement in Direct Preference Optimization was accepted to ICLR 2025.
01/2025
Paper
New paper Provable unlearning in topic modeling and downstream tasks was accepted to ICLR 2025.
01/2025
Paper
New paper Adaptive Data Optimization: Dynamic Sample Selection with Scaling Law was accepted to ICLR 2025.
01/2025
Paper
New paper MUSE: Machine Unlearning Six-Way Evaluation for Language Models was accepted to ICLR 2025.