In an effort to improve generalization in deep learning, we propose SALR: a sharpness-aware learning rate update technique designed to recover flat minimizers. Our method dynamically updates the learning rate of gradient-based optimizers based on the local sharpness of the loss function. This allows optimizers to automatically increase learning rates at sharp valleys to increase the chance of escaping them. We demonstrate the effectiveness of SALR when adopted by various algorithms over a broad range of networks. Our experiments indicate that SALR improves generalization, converges faster, and drives solutions to significantly flatter regions.

Methods for supervised principal component analysis (SPCA) aim to incorporate label information into principal component analysis (PCA), so that the extracted features are more useful for a prediction task of interest. Prior work on SPCA has focused primarily on optimizing prediction error, and has neglected the value of maximizing variance explained by the extracted features. We propose a new method for SPCA that addresses both of these objectives jointly, and demonstrate empirically that our approach dominates existing approaches, i.e., outperforms them with respect to both prediction error and variation explained. Our approach accommodates arbitrary supervised learning losses and, through a statistical reformulation, provides a novel low-rank extension of generalized linear models.

One important goal in sequential decision making problems is to construct a policy that maximizes the average reward over a certain amount of the time. Depending on the purpose of applications, the duration of the learned policy for use in the future (i.e., the planning horizon) is often unknown and can be different from what we consider in the stage of policy optimization. In addition, the performance measure used in learning the policy often depends on the choice of the initial state's distribution. It is always of a great interest to learn a policy with strong generalizability and adaptivity. Given a pre-collected data of multiple trajectories consisting of states, actions and rewards, our goal is to learn a robust policy in the sense that it can guarantee the uniform performance over the unknown planning horizon and the distributional change in the initial state.

Learning-based methodologies increasingly find applications in safety-critical domains like autonomous driving and medical robotics. Due to the rare nature of dangerous events, real-world testing is prohibitively expensive and unscalable. In this work, we employ a probabilistic approach to safety evaluation in simulation, where we are concerned with computing the probability of dangerous events. We develop a novel rare-event simulation method that combines exploration, exploitation, and optimization techniques to find failure modes and estimate their rate of occurrence. We provide rigorous guarantees for the performance of our method in terms of both statistical and computational efficiency. Finally, we demonstrate the efficacy of our approach on a variety of scenarios, illustrating its usefulness as a tool for rapid sensitivity analysis and model comparison that are essential to developing and testing safety-critical autonomous systems.

Optimal Transport is a theory that allows to define geometrical notions of distance between probability distributions and to find correspondences, relationships, between sets of points. Many machine learning applications are derived from this theory, at the frontier between mathematics and optimization. This thesis proposes to study the complex scenario in which the different data belong to incomparable spaces. In particular we address the following questions: how to define and apply Optimal Transport between graphs, between structured data? How can it be adapted when the data are varied and not embedded in the same metric space? This thesis proposes a set of Optimal Transport tools for these different cases. An important part is notably devoted to the study of the Gromov-Wasserstein distance whose properties allow to define interesting transport problems on incomparable spaces. More broadly, we analyze the mathematical properties of the various proposed tools, we establish algorithmic solutions to compute them and we study their applicability in numerous machine learning scenarii which cover, in particular, classification, simplification, partitioning of structured data, as well as heterogeneous domain adaptation.

These authors have contributed equally to this work. Abstract Stationary robotic trainers are lower limb rehab robots which often inco rporate an exoskeleton attached to a stationary base. The issue observed in the stationery trainers for simultaneous knee and ankle - foot complex joints is that they restrict the natura l motion of ankle - foot in the rehab trainings due to the insufficient D e grees of F reedom (DOFs) of these trainers. A new stationary knee - ankle - foot rehab robot with all necessary DOFs is developed here . A typical rehab training is first implemented in simulation, and then tested on a healthy subject. R esults show that the prop osed system functions naturally and meets the requirements of the desired rehab training . I ntroduction Damage to the nervou s system, caused by accidents such as stroke and spinal cord injuries, often lead s to movement disorders [1] . This issue is of practical importance, a s it can severely hamper activities of daily living for the survivors, in light of the large number of stroke incidents per year [2] . To address this issue, patients take rehabilitation exercises under the supervision of therapists to regain their a bilities. During the last two decades, rehabilitation exercises have received sig nificant attention from robotic researchers. Many robots have been designed to facilitate rehab trainings for both patients and therapists, and to improve the training result s in terms of repeatability, reliability and accuracy in evaluating the patient ' s progress [3] . Among the developed platforms, lower limb rehab robots are of more impo rtance, since they have a direct effect on gait. The lower limb rehab robots can be divided into two general types: standing/walking and sitting/lying. The standing/walking robots mainly aim to correct the gait pattern in conditions similar to daily life.

Subset selection algorithms are ubiquitous in AI-driven applications, including, online recruiting portals and image search engines, so it is imperative that these tools are not discriminatory on the basis of protected attributes such as gender or race. Currently, fair subset selection algorithms assume that the protected attributes are known as part of the dataset. However, attributes may be noisy due to errors during data collection or if they are imputed (as is often the case in real-world settings). While a wide body of work addresses the effect of noise on the performance of machine learning algorithms, its effect on fairness remains largely unexamined. We find that in the presence of noisy protected attributes, in attempting to increase fairness without considering noise, one can, in fact, decrease the fairness of the result! Towards addressing this, we consider an existing noise model in which there is probabilistic information about the protected attributes (e.g.,[19, 32, 56, 44]), and ask is fair selection is possible under noisy conditions? We formulate a ``denoised'' selection problem which functions for a large class of fairness metrics; given the desired fairness goal, the solution to the denoised problem violates the goal by at most a small multiplicative amount with high probability. Although the denoised problem turns out to be NP-hard, we give a linear-programming based approximation algorithm for it. We empirically evaluate our approach on both synthetic and real-world datasets. Our empirical results show that this approach can produce subsets which significantly improve the fairness metrics despite the presence of noisy protected attributes, and, compared to prior noise-oblivious approaches, has better Pareto-tradeoffs between utility and fairness.

Domain Adaptation (DA) enables transferring a learning machine from a labeled source domain to an unlabeled target one. While remarkable advances have been made, most of the existing DA methods focus on improving the target accuracy at inference. How to estimate the predictive uncertainty of DA models is vital for decision-making in safety-critical scenarios but remains the boundary to explore. In this paper, we delve into the open problem of Calibration in DA, which is extremely challenging due to the coexistence of domain shift and the lack of target labels. We first reveal the dilemma that DA models learn higher accuracy at the expense of well-calibrated probabilities. Driven by this finding, we propose Transferable Calibration (TransCal) to achieve more accurate calibration with lower bias and variance in a unified hyperparameter-free optimization framework. As a general post-hoc calibration method, TransCal can be easily applied to recalibrate existing DA methods. Its efficacy has been justified both theoretically and empirically.

Min-max optimization of an objective function $f: \mathbb{R}^d \times \mathbb{R}^d \rightarrow \mathbb{R}$ is an important model for robustness in an adversarial setting, with applications to many areas including optimization, economics, and deep learning. In many of these applications $f$ may be nonconvex-nonconcave, and finding a global min-max point may be computationally intractable. There is a long line of work that seeks computationally tractable algorithms for alternatives to the min-max optimization model. However, many of the alternative models have solution points which are only guaranteed to exist under strong assumptions on $f$, such as convexity, monotonicity, or special properties of the starting point. In this paper, we propose an optimization model, the $\varepsilon$-greedy adversarial equilibrium, which can serve as a computationally tractable alternative to the min-max optimization model. Roughly, we say a point $(x^\star, y^\star)$ is an $\varepsilon$-greedy adversarial equilibrium if $y^\star$ is an $\varepsilon$-approximate local maximum for $f(x^\star,\cdot)$, and $x^\star$ is an $\varepsilon$-approximate local minimum for a "greedy approximation" to the function $\max_z f(x, z)$ which can be efficiently estimated using second-order optimization algorithms. The existence follows from an algorithm that converges from any starting point to such a point in a number of evaluations to $f$, $\nabla_{y} f(x,y)$, and $\nabla^2_y f(x,y)$, that is polynomial in $1/\varepsilon$, the dimension $d$, and the bounds on $f$ and its Lipschitz constant. In addition to existence, our model retains many desirable properties of the min-max model. For instance, it empowers the min-player to make updates that take into account the max-player's response, and in the case of strong convexity/concavity it corresponds to a global min-max solution with duality gap $O(\epsilon^2)$.

Simulated annealing (SA) is a probabilistic technique for approximating the global optimum of a given function. Specifically, it is a metaheuristic to approximate global optimization in a large search space for an optimization problem. It is often used when the search space is discrete (e.g., the traveling salesman problem). For problems where finding an approximate global optimum is more important than finding a precise local optimum in a fixed amount of time, simulated annealing may be preferable to exact algorithms such as gradient descent, Branch and Bound. The name of the algorithm comes from annealing in metallurgy, a technique involving heating and controlled cooling of a material to increase the size of its crystals and reduce their defects.