Harvard CMSA Courses

Supercool Composite Dark Matter Beyond 100 TeV

Explore the physics of supercooled confinement transitions in the early universe and their impact on composite dark matter formation, abundance, and cosmological implications.

• YouTube
• 31 minutes
• On-Demand
• Free Video

Charge-Monopole Pairwise Phases from Dressed Quantum States

Explore quantum physics concepts through an analysis of charge-monopole interactions, pairwise phases, and dressed quantum states, focusing on magnetic monopoles and their theoretical implications.

• YouTube
• 31 minutes
• On-Demand
• Free Video

Vandermonde Cells as Positive Geometries

Explore the complexities of Vandermonde cells in positive geometry, focusing on boundary challenges, intersections, and the limitations of canonical forms in semialgebraic subsets of R^n.

• YouTube
• 19 minutes
• On-Demand
• Free Video

Private Multi-Task Learning - Formulation and Applications to Federated Learning

Explore privacy-preserving multi-task learning techniques and their applications in federated learning, focusing on joint differential privacy for protecting sensitive data across healthcare and IoT domains.

• YouTube
• 18 minutes
• On-Demand
• Free Video

Classification of Strategic Agents in Gaming and Self-Improvement Systems

Explore strategic classification systems where agents can both game and genuinely improve, using loan applications as a case study to balance false positives with encouraging authentic improvement.

• YouTube
• 18 minutes
• On-Demand
• Free Video

Bayesian Persuasion for Algorithmic Recourse

Explore how Bayesian persuasion can optimize algorithmic decision-making by providing strategic recommendations while maintaining assessment rule privacy and improving outcomes for all parties involved.

• YouTube
• 17 minutes
• On-Demand
• Free Video

Efficient Mean Estimation with Pure Differential Privacy via Sum-of-Squares Exponential Mechanism

Dive into polynomial-time algorithms for mean estimation using Sum-of-Squares method, exploring pure differential privacy and efficient computational techniques in high-dimensional statistics.

• YouTube
• 18 minutes
• On-Demand
• Free Video

Differentially Private Learning Needs Hidden State or Much Faster Convergence

Explore advanced privacy analysis in machine learning, focusing on hidden-state techniques for differential privacy and their impact on convergence rates in gradient descent algorithms.

• YouTube
• 17 minutes
• On-Demand
• Free Video

Individually-Fair Auctions for Multi-Slot Sponsored Search

Delve into the design of fair-sponsored search auctions, exploring multiplicative click-through rates, preference-based fairness guarantees, and efficient payment computation algorithms for multi-slot settings.

• YouTube
• 19 minutes
• On-Demand
• Free Video

Improved Regret for Differentially Private Exploration in Linear MDP

Delve into privacy-preserving exploration in reinforcement learning, focusing on linear MDP environments and improved regret rates for handling sensitive data in sequential decision-making.

• YouTube
• 18 minutes
• On-Demand
• Free Video

Leximax Approximations and Representative Cohort Selection

Explore mathematical concepts of leximax approximations and their application in selecting representative groups, focusing on algorithmic efficiency and utility improvements in cohort selection.

• YouTube
• 19 minutes
• On-Demand
• Free Video

The Power of Differentially Oblivious Shuffle in Distributed Privacy Mechanisms

Explore the role of differentially oblivious shuffles in distributed privacy mechanisms, focusing on privacy amplification theorems and multi-message protocols for data protection.

• YouTube
• 19 minutes
• On-Demand
• Free Video

Robustness Should Not Be at Odds with Accuracy - A Statistical Learning Theory Perspective

Explore the relationship between adversarial robustness and accuracy in deep learning models, examining how locally adaptive measures can maintain both without compromise.

• YouTube
• 19 minutes
• On-Demand
• Free Video

Multiaccurate Proxies for Downstream Fairness

Explore how to develop fair machine learning models when sensitive demographic data is unavailable during training, using proxy models and multiaccuracy constraints to ensure downstream fairness.

• YouTube
• 18 minutes
• On-Demand
• Free Video

Towards Unification and Robustness of Perturbation and Gradient Based Explanations

Explore the unification of perturbation and gradient-based machine learning explanation methods, examining their convergence, robustness, and practical applications in critical domains.

• YouTube
• 19 minutes
• On-Demand
• Free Video