Misha

I am a PhD student in computer science at Carnegie Mellon University advised by Nina Balcan and Ameet Talwalkar. I study foundations and applications of machine learning, especially meta-learning and algorithm design. My work includes some of the first provable guarantees for gradient-based meta-learning (deep learning methods that "learn-to-learn" using multiple learning tasks as data-points) and end-to-end guarantees for learning-augmented algorithms (algorithms that can take advantage of learned predictions about their instances to improve performance). These results are based on a set of theoretical tools I introduced to extend the idea of surrogate losses from supervised learning to instead upper-bound algorithmic costs. In addition to providing natural measures of task-similarity, this approach often yields effective and practical methods, such as for personalized federated learning and multi-dataset differential privacy. I have also led the push to develop automated machine learning methods for diverse tasks and worked on model compression, neural architecture search, the theory of unsupervised learning, and natural language processing.

 

I am a recipient of the Facebook PhD Fellowship and have interned at Google Research - New York, Microsoft Research - New England, the Lawrence Livermore National Lab, and the Princeton Plasma Physics Lab. Previously, I received an AB in Mathematics and an MSE in Computer Science from Princeton University. For contact information (email), please see my CV.

 
Preprints:

Private Algorithms with Private Predictions.

Kareem Amin, Travis Dick, Mikhail Khodak, Sergei Vassilvitskii.
[arXiv]

Meta-Learning in Games.

Keegan Harris*, Ioannis Anagnostides*, Gabriele Farina, Mikhail Khodak, Zhiwei Steven Wu, Tuomas Sandholm.
[arXiv]

Meta-Learning Adversarial Bandits.

Maria-Florina Balcan, Keegan Harris, Mikhail Khodak, Zhiwei Steven Wu.
[arXiv]

AANG: Automating Auxiliary Learning.

Lucio M. Dery, Paul Michel, Mikhail Khodak, Graham Neubig, Ameet Talwalkar.
[arXiv]

Provably Tuning the ElasticNet Across Instances. To appear in NeurIPS 2022.

Maria-Florina Balcan, Mikhail Khodak, Dravyansh Sharma, Ameet Talwalkar.
[arXiv] [talk]

Efficient Architecture Search for Diverse Tasks. To appear in NeurIPS 2022.

Junhong Shen*, Mikhail Khodak*, Ameet Talwalkar.
[arXiv] [slides] [code] [blog]

Learning Predictions for Algorithms with Predictions. To appear in NeurIPS 2022.

Mikhail Khodak, Maria-Florina Balcan, Ameet Talwalkar, Sergei Vassilvitskii.
[arXiv]

NAS-Bench-360: Benchmarking Neural Architecture Search on Diverse Tasks. To appear in NeurIPS 2022 (Datasets and Benchmarks Track).

Renbo Tu*, Nicholas Roberts*, Mikhail Khodak, Junhong Shen, Frederic Sala, Ameet Talwalkar.
[arXiv] [code] [website]

Selected Papers:

Geometry-Aware Gradient Algorithms for Neural Architecture Search. ICLR 2021.

Liam Li*, Mikhail Khodak*, Maria-Florina Balcan, Ameet Talwalkar.
[paper] [arXiv] [slides] [code] [blog] [talk] [Determined]

Adaptive Gradient-Based Meta-Learning Methods. NeurIPS 2019.

Mikhail Khodak, Maria-Florina Balcan, Ameet Talwalkar.
[paper] [arXiv] [poster] [slides] [code] [blog] [talk]

A Theoretical Analysis of Contrastive Unsupervised Representation Learning. ICML 2019.

Sanjeev Arora, Hrishikesh Khandeparkar, Mikhail Khodak, Orestis Plevrakis, Nikunj Saunshi.
[paper] [arXiv] [poster] [slides] [data] [blog] [talk]

A La Carte Embedding: Cheap but Effective Induction of Semantic Feature Vectors. ACL 2018.

Mikhail Khodak*, Nikunj Saunshi*, Yingyu Liang, Tengyu Ma, Brandon Stewart, Sanjeev Arora.
[paper] [arXiv] [slides] [code] [data] [blog] [talk] [R package]