Recent work [Papyan, Han, and Donoho, 2020] discovered a phenomenon called Neural Collapse (NC) that occurs pervasively in today's deep net training paradigm of driving cross-entropy loss towards zero. In this phenomenon, the last-layer features collapse to their class-means, both the classifiers and class-means collapse to the same Simplex Equiangular Tight Frame (ETF), and the behavior of the last-layer classifier converges to that of the nearest-class-mean decision rule. Since then, follow-ups-such as Mixon et al. [2020] and Poggio and Liao [2020a,b]-formally analyzed this inductive bias by replacing the hard-to-study cross-entropy by the more tractable mean squared error (MSE) loss. But, these works stopped short of demonstrating the empirical reality of MSE-NC on benchmark datasets and canonical networks-as had been done in Papyan, Han, and Donoho [2020] for the cross-entropy loss. In this work, we establish the empirical reality of MSE-NC by reporting experimental observations for three prototypical networks and five canonical datasets with code for reproducing NC. Following this, we develop three main contributions inspired by MSE-NC. Firstly, we show a new theoretical decomposition of the MSE loss into (A) a term assuming the last-layer classifier is exactly the least-squares or Webb and Lowe [1990] classifier and (B) a term capturing the deviation from this least-squares classifier. Secondly, we exhibit experiments on canonical datasets and networks demonstrating that, during training, term-(B) is negligible. This motivates a new theoretical construct: the central path, where the linear classifier stays MSE-optimal-for the given feature activations-throughout the dynamics. Finally, through our study of continually renormalized gradient flow along the central path, we produce closed-form dynamics that predict full Neural Collapse in an unconstrained features model.

We introduce and investigate matrix approximation by decomposition into a sum of radial basis function (RBF) components. An RBF component is a generalization of the outer product between a pair of vectors, where an RBF function replaces the scalar multiplication between individual vector elements. Even though the RBF functions are positive definite, the summation across components is not restricted to convex combinations and allows us to compute the decomposition for any real matrix that is not necessarily symmetric or positive definite. We formulate the problem of seeking such a decomposition as an optimization problem with a nonlinear and non-convex loss function. Several modern versions of the gradient descent method, including their scalable stochastic counterparts, are used to solve this problem. We provide extensive empirical evidence of the effectiveness of the RBF decomposition and that of the gradient-based fitting algorithm. While being conceptually motivated by singular value decomposition (SVD), our proposed nonlinear counterpart outperforms SVD by drastically reducing the memory required to approximate a data matrix with the same $L_2$-error for a wide range of matrix types. For example, it leads to 2 to 10 times memory save for Gaussian noise, graph adjacency matrices, and kernel matrices. Moreover, this proximity-based decomposition can offer additional interpretability in applications that involve, e.g., capturing the inner low-dimensional structure of the data, retaining graph connectivity structure, and preserving the acutance of images.

Active learning methods have recently surged in the literature due to their ability to solve complex structural reliability problems within an affordable computational cost. These methods are designed by adaptively building an inexpensive surrogate of the original limit-state function. Examples of such surrogates include Gaussian process models which have been adopted in many contributions, the most popular ones being the efficient global reliability analysis (EGRA) and the active Kriging Monte Carlo simulation (AK-MCS), two milestone contributions in the field. In this paper, we first conduct a survey of the recent literature, showing that most of the proposed methods actually span from modifying one or more aspects of the two aforementioned methods. We then propose a generalized modular framework to build on-the-fly efficient active learning strategies by combining the following four ingredients or modules: surrogate model, reliability estimation algorithm, learning function and stopping criterion. Using this framework, we devise 39 strategies for the solution of 20 reliability benchmark problems. The results of this extensive benchmark are analyzed under various criteria leading to a synthesized set of recommendations for practitioners. These may be refined with a priori knowledge about the feature of the problem to solve, i.e., dimensionality and magnitude of the failure probability. This benchmark has eventually highlighted the importance of using surrogates in conjunction with sophisticated reliability estimation algorithms as a way to enhance the efficiency of the latter.

Generalized self-concordance is a key property present in the objective function of many important learning problems. We establish the convergence rate of a simple Frank-Wolfe variant that uses the open-loop step size strategy $\gamma_t = 2/(t+2)$, obtaining a $\mathcal{O}(1/t)$ convergence rate for this class of functions in terms of primal gap and Frank-Wolfe gap, where $t$ is the iteration count. This avoids the use of second-order information or the need to estimate local smoothness parameters of previous work. We also show improved convergence rates for various common cases, e.g., when the feasible region under consideration is uniformly convex or polyhedral.

We present a novel approach for the problem of frequency estimation in data streams that is based on optimization and machine learning. Contrary to state-of-the-art streaming frequency estimation algorithms, which heavily rely on random hashing to maintain the frequency distribution of the data steam using limited storage, the proposed approach exploits an observed stream prefix to near-optimally hash elements and compress the target frequency distribution. We develop an exact mixed-integer linear optimization formulation, which enables us to compute optimal or near-optimal hashing schemes for elements seen in the observed stream prefix; then, we use machine learning to hash unseen elements. Further, we develop an efficient block coordinate descent algorithm, which, as we empirically show, produces high quality solutions, and, in a special case, we are able to solve the proposed formulation exactly in linear time using dynamic programming. We empirically evaluate the proposed approach both on synthetic datasets and on real-world search query data. We show that the proposed approach outperforms existing approaches by one to two orders of magnitude in terms of its average (per element) estimation error and by 45-90% in terms of its expected magnitude of estimation error.

Heterogeneous computing systems provide high performance and energy efficiency. However, to optimally utilize such systems, solutions that distribute the work across host CPUs and accelerating devices are needed. In this paper, we present a performance and energy aware approach that combines AI planning heuristics for parameter space exploration with a machine learning model for performance and energy evaluation to determine a near-optimal system configuration. For data-parallel applications our approach determines a near-optimal host-device distribution of work, number of processing units required and the corresponding scheduling strategy. We evaluate our approach for various heterogeneous systems accelerated with GPU or the Intel Xeon Phi. The experimental results demonstrate that our approach finds a near-optimal system configuration by evaluating only about 7% of reasonable configurations. Furthermore, the performance per Joule estimation of system configurations using our machine learning model is more than 1000x faster compared to the system evaluation by program execution.

The goal of Multi-task Bayesian Optimization (MBO) is to minimize the number of queries required to accurately optimize a target black-box function, given access to offline evaluations of other auxiliary functions. When offline datasets are large, the scalability of prior approaches comes at the expense of expressivity and inference quality. We propose JUMBO, an MBO algorithm that sidesteps these limitations by querying additional data based on a combination of acquisition signals derived from training two Gaussian Processes (GP): a cold-GP operating directly in the input domain and a warm-GP that operates in the feature space of a deep neural network pretrained using the offline data. Such a decomposition can dynamically control the reliability of information derived from the online and offline data and the use of pretrained neural networks permits scalability to large offline datasets. Theoretically, we derive regret bounds for JUMBO and show that it achieves no-regret under conditions analogous to GP-UCB (Srinivas et. al. 2010). Empirically, we demonstrate significant performance improvements over existing approaches on two real-world optimization problems: hyper-parameter optimization and automated circuit design.

Iteratively reweighted least square (IRLS) is a popular approach to solve sparsity-enforcing regression problems in machine learning. State of the art approaches are more efficient but typically rely on specific coordinate pruning schemes. In this work, we show how a surprisingly simple reparametrization of IRLS, coupled with a bilevel resolution (instead of an alternating scheme) is able to achieve top performances on a wide range of sparsity (such as Lasso, group Lasso and trace norm regularizations), regularization strength (including hard constraints), and design matrices (ranging from correlated designs to differential operators). Similarly to IRLS, our method only involves linear systems resolutions, but in sharp contrast, corresponds to the minimization of a smooth function. Despite being non-convex, we show that there is no spurious minima and that saddle points are "ridable", so that there always exists a descent direction. We thus advocate for the use of a BFGS quasi-Newton solver, which makes our approach simple, robust and efficient. We perform a numerical benchmark of the convergence speed of our algorithm against state of the art solvers for Lasso, group Lasso, trace norm and linearly constrained problems. These results highlight the versatility of our approach, removing the need to use different solvers depending on the specificity of the ML problem under study.

The Ebola virus and the disease in effect tend to randomly move individuals in the population around susceptible, infected, quarantined, hospitalized, recovered, and dead sub-population. Motivated by the effectiveness in propagating the disease through the virus, a new bio-inspired and population-based optimization algorithm is proposed. This paper presents a novel metaheuristic algorithm named Ebola optimization algorithm (EOSA). To correctly achieve this, this study models the propagation mechanism of the Ebola virus disease, emphasising all consistent states of the propagation. The model was further represented using a mathematical model based on first-order differential equations. After that, the combined propagation and mathematical models were adapted for developing the new metaheuristic algorithm. To evaluate the proposed method's performance and capability compared with other optimization methods, the underlying propagation and mathematical models were first investigated to determine how they successfully simulate the EVD. Furthermore, two sets of benchmark functions consisting of forty-seven (47) classical and over thirty (30) constrained IEEE CEC-2017 benchmark functions are investigated numerically. The results indicate that the performance of the proposed algorithm is competitive with other state-of-the-art optimization methods based on scalability analysis, convergence analysis, and sensitivity analysis. Extensive simulation results indicate that the EOSA outperforms other state-of-the-art popular metaheuristic optimization algorithms such as the Particle Swarm Optimization Algorithm (PSO), Genetic Algorithm (GA), and Artificial Bee Colony Algorithm (ABC) on some shifted, high dimensional and large search range problems.

In this paper, we study the convergence properties of off-policy policy improvement algorithms with state-action density ratio correction under function approximation setting, where the objective function is formulated as a max-max-min optimization problem. We characterize the bias of the learning objective and present two strategies with finite-time convergence guarantees. In our first strategy, we present algorithm P-SREDA with convergence rate $O(\epsilon^{-3})$, whose dependency on $\epsilon$ is optimal. In our second strategy, we propose a new off-policy actor-critic style algorithm named O-SPIM. We prove that O-SPIM converges to a stationary point with total complexity $O(\epsilon^{-4})$, which matches the convergence rate of some recent actor-critic algorithms in the on-policy setting.