Samuel Deng

Hi there! I’m Sam, a PhD candidate in the Theory of Computation Group at Columbia University, where I am extremely fortunate to be advised by Daniel Hsu and Jeannette Wing. My broad research areas are: algorithmic statistics, machine learning theory, and online learning. I was also an undergraduate at Columbia, where I double majored in computer science and philosophy.

A bit more specifically, my research focuses on statistical learning in settings where one cares about learning not just on average over a population, but on a (potentially very large) number of overlapping subgroups of the population. Such multi-group considerations can be captured in formalizations such as multicalibration or multi-group PAC learning, and they are meant to model problems that have more complex desiderata such as fairness or robustness. I also like thinking about online learning, sequential decision-making, and all the cool theory that comes out of it.

I’m grateful to have my research supported by the Avanessians Doctoral Fellowship for Engineering Thought Leaders and Innovators in Data Science and my teaching in the summer of 2024 supported by a SEAS Doctoral Teaching Fellowship. In the Fall of 2024, I was a visiting student at the Simons program on Modern Paradigms of Generalization.

Research

Mathematics for Machine Learning: A Bridge Course
Samuel Deng.
Technical Symposium on Computer Science Education (SIGCSE TS), 2025.
Poster

Group-wise oracle-efficient algorithms for online multi-group learning
Samuel Deng, Daniel Hsu, and Jingwen Liu.
Advances in Neural Information Processing Systems (NeurIPS), 2024. Poster

Multi-group Learning for Hierarchical Groups
Samuel Deng and Daniel Hsu.
International Conference on Machine Learning (ICML), 2024. Poster

Learning Tensor Representations for Meta-Learning.
Samuel Deng, Yilin Guo, Daniel Hsu, Debmalya Mandal.
International Conference on Artificial Intelligence and Statistics (AISTATS), 2022.

A Separation Result Between Data-oblivious and Data-aware Poisoning Attacks.
Samuel Deng, Sanjam Garg, Somesh Jha, Saeed Mahloujifar, Mohammad Mahmoody, Abhradeep Thakurta.
Advances in Neural Information Processing Systems (NeurIPS), 2021.

An Attack on InstaHide: Is Private Learning Possible with Instance Encoding?
Nicholas Carlini, Samuel Deng, Sanjam Garg, Somesh Jha, Saeed Mahloujifar, Mohammad Mahmoody, Shuang Song, Abhradeep Thakurta, Florian Tramèr.
IEEE Symposium on Security and Privacy (Oakland), 2021.

Ensuring Fairness Beyond the Training Data.
Debmalya Mandal, Samuel Deng, Suman Jana, Jeannette Wing, Daniel Hsu.
Advances in Neural Information Processing Systems (NeurIPS), 2020.

Biased Programmers? Or Biased Data? A Field Experiment on Operationalizing AI Ethics.
Bo Cowgill, Fabrizio Dell’Acqua, Samuel Deng, Daniel Hsu, Nakul Verma, Augustin Chaintreau.
21st ACM Conference on Economics and Computation, 2020.

Methodological Blind Spots in Machine Learning Fairness: Lessons from the Philosophy of Science and Computer Science
Samuel Deng, Achille Varzi.
NeurIPS Workshop on Human-Centric Machine Learning, 2019.
Undergraduate Senior Thesis, 2019. full pdf

Teaching

Teaching Philosophy. I really love teaching, and I am passionate about continuously developing as a teacher. My teaching philosophy centers around three core principles:

A driving and cohesive narrative should propel all parts of a course.
Ideas should be presented as if the student could’ve discovered them themselves.
An instructor should never forget how they first struggled when learning the same ideas.

I’ve also constructed a draft teaching portfolio that compiles all the feedback I’ve received thus far on my teaching and dives into how I try to practice this philosophy with several representative artifacts. This is a work in progress; it’ll eventually be another part of my website (not a clunky 40 MB document!)

Some Teaching Experience. In Summer 2024, I created and taught Mathematics for Machine Learning from scratch, a bridge course for Columbia CS students to strengthen mathematical foundations for studying machine learning. See the link for the course materials, which are all public. Here’s a bit on why I made this course (tldr: when I took machine learning, I had no idea what an expectation was). The course has since been added to Columbia’s official Computer Science curriculum, and I presented a poster on this course at SIGCSE TS 2025. I will be teaching this again in Summer 2025.

On the teaching front, I’ve also had the pleasure of:

Participating in Columbia’s Center for Teaching and Learning (CTL) Teaching Development Program, a multiyear certification program for students to cultivate, document, and reflect on the development of their teaching throughout their Ph.D.
Serving as Head TA for Daniel Hsu’s Computational Linear Algebra, Fall 2022. (course webpage) Delivered a guest lecture on eigenvectors and eigenvalues as part of an observation for CTL’s Foundation Track program (12/01/2022), as well as weekly recitations.
Serving as an Instructor for Christos Papadimitriou and John Morrison’s Natural and Artificial Neural Networks Lab, Spring 2022. My co-instructor, Clayton Sanford, and I designed the companion course from scratch: all materials are available here.
Serving as a Teaching Assistant Fellow and Head Teaching Assistant during my M.S. for Machine Learning and Discrete Mathematics, where I was awarded the Andrew P. Kosoresow Award for Excellence in Teaching.
Serving as a TA for Machine Learning and Discrete Mathematics as an undergraduate at Columbia.

Service

Alongside Hadleigh Schwartz, I am currently Ph.D coordinator for Columbia’s Emerging Scholars Program (ESP), a peer-taught, discussion-based seminar course for first and second-year CS students focused on group problem-solving, collaboration, and introducing beginning computer scientists to the breadth of the subject. Please reach out if you’d like to learn more about this program!

For the broader scientific community, I have also served as a reviewer for: NeurIPS 2024 (Top Reviewer), ICLR 2025, ICML 2025.

Miscellaneous

I also like long-distance running, fiddling around poorly on the guitar, nerding out about b o o k s and mo vi es , and a good burrito.