• Adversarial KA

    Virtual

    Freedman CMSA Seminar Speaker: Slava Dzhenzher, MIPT Title: Adversarial KA Abstract: Regarding the representation theorem of Kolmogorov and Arnold (KA) as an algorithm for representing or «expressing» functions, we test its robustness by analyzing its ability to withstand adversarial attacks. We find KA to be robust to countable collections of continuous adversaries, but unearth a […]

  • Bass-Note Spectra of locally uniform geometries

    CMSA Room G10 CMSA, 20 Garden Street, Cambridge, MA, United States

    Colloquium Speaker: Peter Sarnak, IAS & Princeton University Title: Bass-Note Spectra of locally uniform geometries Abstract: We formulate and report on the problem of the Bass-Note Spectrum of an invariant operator as one varies over locally uniform geometries. In the Euclidean setting this recasts classical problems of Mahler from the geometry of numbers in a new […]

  • Algebraic approach to the bow construction

    Science Center 507 1 Oxford Street, Cambridge

    Geometry and Quantum Theory Seminar Speakers: Anadil Saeed Rao (Northeastern) and Vasily Krylov (CMSA/Harvard) Title: Algebraic approach to the bow construction Abstract: Anadil will finish his part of the talk. Vasily will then explain how to understand moduli spaces of instantons on R^4, monopoles on R^3, and instantons on Taub-NUT spaces algebraically as so-called bow varieties. We […]

  • Incentives for data sharing in federated learning

    CMSA Room G10 CMSA, 20 Garden Street, Cambridge, MA, United States

    Member Seminar Speaker: Han Shao, Harvard CMSA Title: Incentives for data sharing in federated learning Abstract: Federated learning has recently emerged as a powerful approach for enabling collaboration across large populations of learning agents. However, agents may have incentives to defect from the collaboration—that is, to withdraw or contribute less data than expected—due to the […]

  • Thinking Outside the Ballot Box

    CMSA Room G10 CMSA, 20 Garden Street, Cambridge, MA, United States

    Colloquium Speaker: Ariel Procaccia, Harvard University Title: Thinking Outside the Ballot Box Abstract: How should one design unprecedented democratic processes capable of handling enormous sets of alternatives like all possible policies, bills, or statements? I argue that this challenge can be addressed through a framework called generative social choice, which fuses the rigor of social choice theory […]

  • Asset pricing with heterogeneous agents

    CMSA Room G10 CMSA, 20 Garden Street, Cambridge, MA, United States

    Member Seminar Speaker: Sergiy Verstyuk, Harvard CMSA Title: Asset pricing with heterogeneous agents Abstract: This talk will introduce the basics of continuous-time finance, discuss important existing theories and models, as well as present some new asset pricing results in a setting with many heterogeneous investors. (Joint work with Puskar Mondal.)  

  • Factorizations for data analysis

    CMSA Room G10 CMSA, 20 Garden Street, Cambridge, MA, United States

    Colloquium Speaker: Anna Seigal, Harvard University Title: Factorizations for data analysis Abstract: We can find structure in data by factoring it into building blocks, which should be interpretable for the context at hand. A classical example is principal component analysis (PCA), which uses the eigendecomposition of the covariance matrix to find axes of variation in a dataset. Starting from […]

  • Report on the Perimeter Institute Theory+AI Workshop

    CMSA 20 Garden Street Cambridge, Massachusetts 02138 United States

    Conference Reports Speaker: Hugo Cui, Harvard CMSA Title: Report on the Perimeter Institute Theory+AI Workshop Abstract: I will give a survey and brief summary of some of the talks given at the Theory+AI Workshop: Theoretical Physics for AI event organized by Perimeter Institute in April, on approaches to machine learning theory inspired from physics. Link : https://events.perimeterinstitute.ca/event/993/

  • Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining

    CMSA Room G10 CMSA, 20 Garden Street, Cambridge, MA, United States

    Member Seminar Speaker: Samy Jelassi, CMSA Title: Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining Abstract: Reinforcement Learning has become a crucial step in training state-of-the-art language models such as DeepSeek-R1 for solving mathematical problems. In this talk, I will first review the mechanisms of Reinforcement Learning fine-tuning. Then, I will present a systematic end-to-end […]