Her research interests focus on using data to find insights that can be turned into learning interventions. Bart Workshop Materials Logo credit: Convergence in routing games and beyond. The Hedge Algorithm on a Continuum. However, convergence of the actual sequence is not guaranteed in general. We collect a dataset using this platform, then apply the proposed method to estimate the learning rates of each player. The onramp dynamics is modeled using an ordinary differential equation describing the evolution of the queue length.

A greedy algorithm and a mirror descent algorithm based on the adjoint method. Finally, we propose a generalization to the dual averaging method on the set of Lebesgue-continuous distributions over S. Adjoint-based optimization on a network of discretized scalar conservation law PDEs with applications to coordinated ramp metering. And then, I present my deployment of a scaled hint intervention using the insights from the analysis. We compare the performance of these methods in terms of achieved cost and computational complexity on parallel networks, and on a model of the Los Angeles highway network. The Hedge Algorithm on a Continuum.

A new class of latency functions is introduced to model congestion due to the formation of physical queues, inspired from the fundamental diagram of traffic.

Convergence, Estimation of Player Dyanmics, and Control. We consider a model in which players use regret-minimizing algorithms as the learning mechanism, and study the resulting dynamics. Next, I discuss analyzing a large data set of constructed-response, code-tracing wrong answers using mixed methods of quantitative and qualitative techniques.

We provide a simple polynomial-time algorithm for computing the best Nash equilibrium, i. In particular, we derive the adjoint system equations of the Hedge dynamics and show that they can be solved efficiently. Repeated routing, online learning, and no-regret algorithms. The game is stochastic in that each player observes a stochastic vector, the conditional expectation of which is equal to the true loss almost surely.

We present the result of some simulations and numerically check the convergence of the method. In particular, we give an averaging interpretation of accelerated dynamics, and derive simple sufficient conditions on the averaging scheme to guarantee a given rate of convergence.

For this new class, some results from the classical congestion games literature in which latency is assumed to be a non-decreasing function of the flow do not hold. Online learning and convex optimization algorithms have become essential tools for solving gerkeley in modern machine learning, berkelej and engineering. A simple Stackelberg strategy, the non-compliant first NCF strategy, is introduced, which can be computed in polynomial time, and it is shown to be optimal for tak new class of latency on parallel networks.

The method is applied to the problem of coordinated ramp metering on freeway networks. Classes are growing in size and adding more and more technology. In doing so, we first prove that if both players play a Hannan-consistent strategy, then with probability 1 the empirical distributions of play weakly converge to the set of Nash equilibria of the game.

We also prove a general lower bound on the worst-case regret for any online algorithm. First, we introduce a new class of latency functions that models ttalk due to the formation of physical queues. The shuffling leads to reduced image blur at the cost of noise-like artifacts. Minimizing regret on reflexive Banach spaces and Nash equilibria in continuous zero-sum games. We study the problem of learning similarity functions over very large corpora using neural network embedding models.

## Kate Harrison

Chua Award for dissegtation achievement in nonlinear science. No-regret learning algorithms are known to guarantee convergence of a subsequence of population strategies. By studying the spectrum of the linearized system around rest points, we show that Nash equilibria are locally asymptotically stable stationary points.

This is motivated by the fact that this spatiotemporal information can easily be used as the basis for inferences for a person’s activities. I was awarded the Leon O.

Numerical simulations on the I15 freeway in California demonstrate an improvement in performance and running time compared with existing methods. This is a common problem in first-order methods for convex optimization and online-learning algorithms, such as mirror descent.

This results in an optimal control problem under learning dynamics. We show that the gradient of this term can be efficiently computed by maintaining estimates of the Gramians, and develop variance reduction schemes to improve the quality of the estimates. Efficient Bregman Projections onto the Simplex. We discuss the interaction between the parameters of the dynamics learning rate and averaging rates and the covariation of the noise process.

ICML On the convergence of online learning in selfish routing.

# Jon Tamir – Home

The experiments indicate that adaptive averaging performs at least as well as adaptive restarting, with significant improvements in some cases.

Benjamin received the Grand Prix d’option of Ecole Polytechnique.

Then, by carefully discretizing the ODE, we obtain a family of accelerated algorithms with optimal rate of convergence.