Improving the performance of deep networks in data-limited regimes has warranted much attention. In this work, we empirically show that "winning tickets" (small subnetworks) obtained via magnitude pruning based on the lottery ticket hypothesis [1], apart from being sparse are also effective recognizers in data-limited regimes. Based on extensive experiments, we find that in low data regimes (datasets of 50-100 examples per class), sparse winning tickets substantially outperform the original dense networks. This approach, when combined with augmentations or fine-tuning from a self-supervised backbone network, shows further improvements in performance by as much as 16% (absolute) on low sample datasets and longtailed classification. Further, sparse winning tickets are more robust to synthetic noise and distribution shifts compared to their dense counterparts. Our analysis of winning tickets on small datasets indicates that, though sparse, the networks retain density in the initial layers and their representations are more generalizable.

When solving two-player zero-sum games, multi-agent reinforcement learning (MARL) algorithms often create populations of agents where, at each iteration, a new agent is discovered as the best response to a mixture over the opponent population. Within such a process, the update rules of "who to compete with" (i.e., the opponent mixture) and "how to beat them" (i.e., finding best responses) are underpinned by manually developed game theoretical principles such as fictitious play and Double Oracle. In this paper1, we introduce a novel framework--Neural Auto-Curricula (NAC)--that leverages meta-gradient descent to automate the discovery of the learning update rule without explicit human design. Specifically, we parameterise the opponent selection module by neural networks and the bestresponse module by optimisation subroutines, and update their parameters solely via interaction with the game engine, where both players aim to minimise their exploitability. Surprisingly, even without human design, the discovered MARL algorithms achieve competitive or even better performance with the state-of-the-art population-based game solvers (e.g., PSRO) on Games of Skill, differentiable Lotto, non-transitive Mixture Games, Iterated Matching Pennies, and Kuhn Poker. Additionally, we show that NAC is able to generalise from small games to large games, for example training on Kuhn Poker and outperforming PSRO on Leduc Poker. Our work inspires a promising future direction to discover general MARL algorithms solely from data.

Algorithm 1 Hard-gating Algorithm for In-Stage IDKCascade Input Ds: Training data containing Ns samples in stage-s Ms: Sorted list of the models trained for stage-s C: Dictionary of models' spatio-temporal costs cs: User-defined budget of spatio-temporal cost for stage-s q: Confidence function maxA: Value for the upper bound of the cutoffs to avoid over-fitting nBins: Number of bins for the grid search Output s: The optimal IDK cutoff vector for stage-s 1: procedure HARDGATING(Ds, Ms, cs, C, q, maxA, nBins) 2: s =[], ModelAssign = 1, cost = P We use the Sepsis-3 toolkit3 to obtain the suspected infection time in patients, and following the process in Seymour et al. (2016) to finally label the onset of sepsis. We result at a total number of 20,009 sepsis patients out of the 52,902 adult patients from MIMIC-III database. We exclude those patients who stay in ICUs less than 6 hours and also exclude those patients who developed sepsis within the first 6 hours after ICU admission. This reduces our cohort to a total of 34,475ICU patient, and only 2,370(6.8%) Then according to Singer et al. (2016), we identify the onset of septic shock as Algorithm 3 End-to-End Training algorithm for UnfoldML Input D: Full training data containing N instances M: Full model zoo C: Dictionary of models' spatio-temporal costs q: Confidence criterion Output: the optimal ICK1 gate parameters (or a,b): the optimal IDK gate parameters 1: procedure END-TO-ENDTRAINING (D, M) 2: Pre-allocate costs cs for each stage s. Figure 4: Transitions in model calls: both cascades always call the first model per each stage for an entrance and transition to next models (IDK) or next stage (ICK).

Self-supervised learning (SSL) methods aim to exploit the abundance of unlabelled data for machine learning (ML), however the underlying principles are often method-specific. An SSL framework derived from biological first principles of embodied learning could unify the various SSL methods, help elucidate learning in the brain, and possibly improve ML. SSL commonly transforms each training datapoint into a pair of views, uses the knowledge of this pairing as a positive (i.e.

We study online change point detection problems under the constraint of local differential privacy (LDP) where, in particular, the statistician does not have access to the raw data. As a concrete problem, we study a multivariate nonparametric regression problem. At each time point t, the raw data are assumed to be of the form (Xt,Yt), where Xt is a d-dimensional feature vector and Yt is a response variable. Our primary aim is to detect changes in the regression function mt(x) = E(Yt|Xt = x) as soon as the change occurs. We provide algorithms which respect the LDP constraint, which control the false alarm probability, and which detect changes with a minimal (minimax rate-optimal) delay. To quantify the cost of privacy, we also present the optimal rate in the benchmark, non-private setting. These non-private results are also new to the literature and thus are interesting per se. In addition, we study the univariate mean online change point detection problem, under privacy constraints. This serves as the blueprint of studying more complicated private change point detection problems.