Andrej Risteski
Associate Professor
Studies theory of machine learning, including unsupervised learning, generative models and deep learning.
Andrej Risteski's lab at CMU works on learning theory & optimization and generative & foundation models. 20 papers since 2023, 18 at top venues, mostly ICLR and ICML. 5 PhD students. 12 alumni, most often to Google. Applications close .
At a glance
- papers, 2023–2026
- 5/ year
- at T1 venues
- 90%
- PhD students
- 5
- MS students
- 0
- alumni listed
- 12
Papers by yearfrom the lab homepage and DBLP, 2023–2026
- 4 papers2026 · T1 2 · collecting
- 4 papers2025 · T1 4
- 4 papers2024 · T1 4
- 8 papers2023 · T1 8
T1 venuesOtherVenues: ICLR, ICML, NeurIPS, COLT
Research focuspaper titles that use each topic's words
- Generative & foundation models3 papers
- Learning theory & optimization2 papers
- studies theory0 papers
- unsupervised learning0 papers
- unsupervised0 papers
Paper list
From the lab homepage and DBLP.
20264 papers
- Trans. Mach. Learn. Res.CodePDE: An Inference Framework for LLM-driven PDE Solver Generation (opens in a new tab)
- COLT (T1)Steering diffusion models with quadratic rewards: a fine-grained analysis (opens in a new tab)
- ICLR (T1)Taming Imperfect Process Verifiers: A Sampling Perspective on Backtracking (opens in a new tab)
- Trans. Mach. Learn. Res.Towards Understanding the Transferability of Adversarial Suffixes in Large Language Models (opens in a new tab)
20254 papers
- ICLR (T1)On the Benefits of Memory for Modeling Time-Dependent PDEs (opens in a new tab)
- ICML (T1)On the Query Complexity of Verifier-Assisted Language Generation (opens in a new tab)
- ICLR (T1)Progressive distillation induces an implicit curriculum (opens in a new tab)
- ICML (T1)Towards characterizing the value of edge embeddings in Graph Neural Networks (opens in a new tab)
20244 papers
- COLT (T1)Fit Like You Sample: Sample-Efficient Generalized Score Matching from Fast Mixing Diffusions (opens in a new tab)
- ICLR (T1)Outliers with Opposing Signals Have an Outsized Effect on Neural Network Optimization (opens in a new tab)
- ICML (T1)Promises and Pitfalls of Generative Masked Language Modeling: Theoretical Framework and Practical Guidelines (opens in a new tab)
- ICLR (T1)Understanding Augmentation-based Self-Supervised Representation Learning via RKHS Approximation and Regression (opens in a new tab)
20238 papers
- NeurIPS (T1)Deep Equilibrium Based Neural Operators for Steady-State PDEs (opens in a new tab)
- ICML (T1)How Do Transformers Learn Topic Structure: Towards a Mechanistic Understanding (opens in a new tab)
- ICML (T1)Neural Network Approximations of PDEs Beyond Linearity: A Representational Perspective (opens in a new tab)
- ICLR (T1)Pitfalls of Gaussians as a noise distribution in NCE (opens in a new tab)
- NeurIPS (T1)Provable benefits of annealing for estimating normalizing constants: Importance Sampling, Noise-Contrastive Estimation, and beyond (opens in a new tab)
- NeurIPS (T1)Provable benefits of score matching (opens in a new tab)
- ICLR (T1)Statistical Efficiency of Score Matching: The View from Isoperimetry (opens in a new tab)
- NeurIPS (T1)Transformers are uninterpretable with myopic methods: a case study with bounded Dyck grammars (opens in a new tab)
Where alumni went12 listed, 17% to industry
Industry 1Academia 5
Most often to
Also to
- Stanford University
- MIT
- the Flatiron Institute and Polymathic AI
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
- USUC Berkeley
Alex Dimakis
- Signal processing & information theory
- USUW
Natasha Jaques
- Reinforcement learning
- CNTsinghua
Xianyuan Zhan
- Reinforcement learning
- USMIT
Ankur Moitra
- Trustworthy & safe ML
- CAToronto
Murat Erdogdu
- Learning theory & optimization
- USUC Berkeley
Jason Lee
- Reinforcement learning
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage and DBLP
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
CMUcloses