Aditi Raghunathan
Assistant Professor
Studies robust and reliable machine learning, including distribution shift, adversarial robustness and adapting foundation models.
Aditi Raghunathan's lab at CMU works on trustworthy & safe ML, generative & foundation models and large language models. 47 papers since 2023, 34 at top venues, mostly ICLR and ICML. 6 PhD students. 10 alumni. Applications close .
At a glance
- papers, 2023–2026
- 12/ year
- at T1 venues
- 72%
- PhD students
- 6
- MS students
- 0
- alumni listed
- 10
Papers by yearfrom the lab homepage, 2023–2026
- 15 papers2026 · T1 9 · collecting
- 16 papers2025 · T1 10
- 10 papers2024 · T1 9
- 6 papers2023 · T1 6
T1 venuesOtherVenues: ICLR, ICML, NeurIPS, CVPR, Findings of ACL
Research focuspaper titles that use each topic's words
- Large language models18 papers
- Generative & foundation models9 papers
- Trustworthy & safe ML1 paper
- shift1 paper
- studies robust0 papers
Paper list
From the lab homepage.
202615 papers
- ICML (T1)Annotations Mitigate Post-training Mode Collapse
- ICML (T1)Demystifying Entropy Control in LLM RL Training: Theoretical Analysis and Dynamic Scheduling
- ICML (T1)Differential Smoothing Mitigates Sharpening and Improves LLM Reasoning
- COLMEarly Data Exposure Improves Robustness to Subsequent Fine-Tuning
- SATMLExact Unlearning of Finetuning Data via Model Merging at Scale
- NeurIPS (T1)Flow Map Language Models: One-step Language Modeling via Continuous Denoising
- ICLR (T1)ImpossibleBench: Measuring LLMs' Propensity of Exploiting Test Cases
- ICLR (T1)Mode-Conditioning Unlocks Superior Test-Time Scaling
- COLMPando: Do Interpretability Methods Work When Models Won't Explain Themselves?
- MOSS Workshop at COLMPretraining Data Scale Reverses the AdamW–Muon Fine-Tuning Gap
- NeurIPS (T1)Self-Trained Verification for Training- and Test-Time Self-Improvement
- ICML (T1)Sharpness-Aware Pretraining Mitigates Catastrophic Forgetting
202516 papers
- ICLR (T1)Context-Parametric Inversion: Why Instruction Finetuning May Not Actually Improve Context Reliance
- ICLR (T1)Dissecting Adversarial Robustness of Multimodal Agents
- ICML (T1)Memorization Sinks: Isolating memorization during LLM training
- Findings of ACLMitigating Bias in RAG: Controlling the Embedder
- COLMMitigating Modal Imbalance in Multimodal Reasoning
- EMNLP (T1)Not-Just-Scaling Laws: Towards a Better Understanding of the Downstream Impact of Language Model Design Decisions
- Findings of NAACLOn the Feasibility of In-Context Probing for Data Attribution
- ICML (T1)Overtrained Language Models Are Harder to Fine-Tune
- NeurIPS (T1)Reasoning as an Adaptive Defense for Safety
- ICLR (T1)Repetition Improves Language Model Embeddings
- ICML (T1)Roll the dice & look before you leap: Going beyond the creative limits of next-token prediction
- ICLR (T1)Scaling Laws for Precision
202410 papers
- TMLR (T1)Multitask Learning Can Improve Worst-Group Outcomes
- NeurIPS (T1)Predicting the Performance of Foundation Models via Agreement-on-the-Line
- CVPR (T1)Scaling Laws for Data Filtering: Data Curation Cannot Be Compute Agnostic
- ICLR (T1)Sharpness-Aware Minimization Enhances Feature Quality via Balanced Learning
- ICLR (T1)T-MARS: Improving Visual Representations by Circumventing Text Feature Learning
- NeurIPS (T1)Test-Time Adaptation Induces Stronger Accuracy and Agreement-on-the-Line
- NeurIPS Safe Generative AI WorkshopTesting the Limits of Jailbreaking Defenses via the Purple Problem
- ICLR (T1)Understanding Catastrophic Forgetting in Language Models via Implicit Inference
- ICML (T1)Understanding Finetuning for Factual Knowledge Extraction
- ICLR (T1)Why is Sharpness-Aware Minimization Robust to Label Noise?
20236 papers
- ICML (T1)Automatically Auditing Large Language Models via Discrete Optimization
- ICLR (T1)Bitrate-Constrained DRO: Beyond Worst Case Robustness To Unknown Group Shifts
- NeurIPS (T1)Complementary Benefits of Contrastive Learning and Self-Training Under Distribution Shift
- ICML (T1)Contextual Reliability: When Different Features Matter in Different Contexts
- CVPR (T1)Finetune Like You Pretrain: Improved Finetuning of Zero-Shot Vision Models
- ICLR (T1)Using Language to Extend to Unseen Domains
Where alumni went10 listed, 0% to industry
Industry 0Academia 10
Also to
- Carnegie Mellon University
- Harvard University
- Stanford University
Apply to CMUFall 2027
CMU SCS Computer Science PhD
- Round
- Single round
- Opens
- GRE
- See program page
- English
- TOEFL / IELTS / Duolingo (F-1/J-1 non-native speakers)
- Fee
- USD 100 (USD 80 by early deadline)
Confirmed on the program page.
Similar labsnearest by research summary
- USNYU
Pavel Izmailov
- Trustworthy & safe ML
- SGNTU
Jaehong Yoon
- Vision-language & multimodal
- USCMU
J. Zico Kolter
- Trustworthy & safe ML
- CNSJTU
Shiyu Liang
- Learning theory & optimization
- USUCSD
Rahul Parhi
- Learning theory & optimization
- GBOxford
Varun Kanade
- Learning theory & optimization
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
CMUcloses