In selection processes such as hiring, promotion, and college admissions, implicit bias toward socially-salient attributes such as race, gender, or sexual orientation of candidates is known to produce persistent inequality and reduce aggregate utility for the decision maker. Interventions such as the Rooney Rule and its generalizations, which require the decision maker to select at least a specified number of individuals from each affected group, have been proposed to mitigate the adverse effects of implicit bias in selection. Recent works have established that such lower-bound constraints can be very effective in improving aggregate utility in the case when each individual belongs to at most one affected group. However, in several settings, individuals may belong to multiple affected groups and, consequently, face more extreme implicit bias due to this intersectionality. We consider independently drawn utilities and show that, in the intersectional case, the aforementioned non-intersectional constraints can only recover part of the total utility achievable in the absence of implicit bias. On the other hand, we show that if one includes appropriate lower-bound constraints on the intersections, almost all the utility achievable in the absence of implicit bias can be recovered. Thus, intersectional constraints can offer a significant advantage over a reductionist dimension-by-dimension non-intersectional approach to reducing inequality.

We study algorithms for learning low-rank neural networks -- networks where the weight parameters are re-parameterized by products of two low-rank matrices. First, we present a provably efficient algorithm which learns an optimal low-rank approximation to a single-hidden-layer ReLU network up to additive error $\epsilon$ with probability $\ge 1 - \delta$, given access to noiseless samples with Gaussian marginals in polynomial time and samples. Thus, we provide the first example of an algorithm which can efficiently learn a neural network up to additive error without assuming the ground truth is realizable. To solve this problem, we introduce an efficient SVD-based $\textit{Nonlinear Kernel Projection}$ algorithm for solving a nonlinear low-rank approximation problem over Gaussian space. Inspired by the efficiency of our algorithm, we propose a novel low-rank initialization framework for training low-rank $\textit{deep}$ networks, and prove that for ReLU networks, the gap between our method and existing schemes widens as the desired rank of the approximating weights decreases, or as the dimension of the inputs increases (the latter point holds when network width is superlinear in dimension). Finally, we validate our theory by training ResNet and EfficientNet models on ImageNet.

Writing about A.I. has given me the opportunity to think more about datascience, programming and technology trends for new professionals. Picture this, a young woman in Africa starts her her own startup to meet unmet needs, the year is 2030. Will she need to even hire developers or can her small IT team handle the engineering of the product? Are we heading to a world where the software revolution will truly be democratized? This is an Op-ed about an emerging and somewhat speculative field that is showing real signs of potential.

Keyword localisation is the task of finding where in a speech utterance a given query keyword occurs. We investigate to what extent keyword localisation is possible using a visually grounded speech (VGS) model. VGS models are trained on unlabelled images paired with spoken captions. These models are therefore self-supervised -- trained without any explicit textual label or location information. To obtain training targets, we first tag training images with soft text labels using a pretrained visual classifier with a fixed vocabulary. This enables a VGS model to predict the presence of a written keyword in an utterance, but not its location. We consider four ways to equip VGS models with localisations capabilities. Two of these -- a saliency approach and input masking -- can be applied to an arbitrary prediction model after training, while the other two -- attention and a score aggregation approach -- are incorporated directly into the structure of the model. Masked-based localisation gives some of the best reported localisation scores from a VGS model, with an accuracy of 57% when the system knows that a keyword occurs in an utterance and need to predict its location. In a setting where localisation is performed after detection, an $F_1$ of 25% is achieved, and in a setting where a keyword spotting ranking pass is first performed, we get a localisation P@10 of 32%. While these scores are modest compared to the idealised setting with unordered bag-of-word-supervision (from transcriptions), these models do not receive any textual or location supervision. Further analyses show that these models are limited by the first detection or ranking pass. Moreover, individual keyword localisation performance is correlated with the tagging performance from the visual classifier. We also show qualitatively how and where semantic mistakes occur, e.g. that the model locates surfer when queried with ocean.

In contrast to SGD, adaptive gradient methods like Adam allow robust training of modern deep networks, especially large language models. However, the use of adaptivity not only comes at the cost of extra memory but also raises the fundamental question: can non-adaptive methods like SGD enjoy similar benefits? In this paper, we provide an affirmative answer to this question by proposing to achieve both robust and memory-efficient training via the following general recipe: (1) modify the architecture and make it scale invariant, i.e. the scale of parameter doesn't affect the output of the network, (2) train with SGD and weight decay, and optionally (3) clip the global gradient norm proportional to weight norm multiplied by $\sqrt{\tfrac{2\lambda}{\eta}}$, where $\eta$ is learning rate and $\lambda$ is weight decay. We show that this general approach is robust to rescaling of parameter and loss by proving that its convergence only depends logarithmically on the scale of initialization and loss, whereas the standard SGD might not even converge for many initializations. Following our recipe, we design a scale invariant version of BERT, called SIBERT, which when trained simply by vanilla SGD achieves performance comparable to BERT trained by adaptive methods like Adam on downstream tasks.

Artificial intelligence (AI) has been increasingly applied in scientific activities for decades; however, it is still far from an insightful and trustworthy collaborator in the scientific process. Most existing AI methods are either too simplistic to be useful in real problems faced by scientists or too domain-specialized (even dogmatized), stifling transformative discoveries or paradigm shifts. We present an AI research associate for early-stage scientific discovery based on (a) a novel minimally-biased ontology for physics-based modeling that is context-aware, interpretable, and generalizable across classical and relativistic physics; (b) automatic search for viable and parsimonious hypotheses, represented at a high-level (via domain-agnostic constructs) with built-in invariants, e.g., postulated forms of conservation principles implied by a presupposed spacetime topology; and (c) automatic compilation of the enumerated hypotheses to domain-specific, interpretable, and trainable/testable tensor-based computation graphs to learn phenomenological relations, e.g., constitutive or material laws, from sparse (and possibly noisy) data sets.

Though there is a strong consensus that word length and frequency are the most important single-word features determining visual-orthographic access to the mental lexicon, there is less agreement as how to best capture syntactic and semantic factors. The traditional approach in cognitive reading research assumes that word predictability from sentence context is best captured by cloze completion probability (CCP) derived from human performance data. We review recent research suggesting that probabilistic language models provide deeper explanations for syntactic and semantic effects than CCP. Then we compare CCP with (1) Symbolic n-gram models consolidate syntactic and semantic short-range relations by computing the probability of a word to occur, given two preceding words. (2) Topic models rely on subsymbolic representations to capture long-range semantic similarity by word co-occurrence counts in documents. (3) In recurrent neural networks (RNNs), the subsymbolic units are trained to predict the next word, given all preceding words in the sentences. To examine lexical retrieval, these models were used to predict single fixation durations and gaze durations to capture rapidly successful and standard lexical access, and total viewing time to capture late semantic integration. The linear item-level analyses showed greater correlations of all language models with all eye-movement measures than CCP. Then we examined non-linear relations between the different types of predictability and the reading times using generalized additive models. N-gram and RNN probabilities of the present word more consistently predicted reading performance compared with topic models or CCP.

Weighted knowledge bases for description logics with typicality have been recently considered under a "concept-wise" multipreference semantics (in both the two-valued and fuzzy case), as the basis of a logical semantics of MultiLayer Perceptrons (MLPs). In this paper we consider weighted conditional ALC knowledge bases with typicality in the finitely many-valued case, through three different semantic constructions, based on coherent, faithful and phi-coherent interpretations. For the boolean fragment LC of ALC we exploit ASP and "asprin" for reasoning with the concept-wise multipreference entailment under a phi-coherent semantics, suitable to characterize the stationary states of MLPs. As a proof of concept, we experiment the proposed approach for checking properties of trained MLPs.

Large-scale deployments of low Earth orbit (LEO) satellites collect massive amount of Earth imageries and sensor data, which can empower machine learning (ML) to address global challenges such as real-time disaster navigation and mitigation. However, it is often infeasible to download all the high-resolution images and train these ML models on the ground because of limited downlink bandwidth, sparse connectivity, and regularization constraints on the imagery resolution. To address these challenges, we leverage Federated Learning (FL), where ground stations and satellites collaboratively train a global ML model without sharing the captured images on the satellites. We show fundamental challenges in applying existing FL algorithms among satellites and ground stations, and we formulate an optimization problem which captures a unique trade-off between staleness and idleness. We propose a novel FL framework, named FedSpace, which dynamically schedules model aggregation based on the deterministic and time-varying connectivity according to satellite orbits. Extensive numerical evaluations based on real-world satellite images and satellite networks show that FedSpace reduces the training time by 1.7 days (38.6%) over the state-of-the-art FL algorithms.

For training neural networks, flat-minima optimizers that seek to find parameters in neighborhoods having uniformly low loss (flat minima) have been shown to improve upon stochastic and adaptive gradient-based methods. Two methods for finding flat minima stand out: 1. Averaging methods (i.e., Stochastic Weight Averaging, SWA), and 2. Minimax methods (i.e., Sharpness Aware Minimization, SAM). However, despite similar motivations, there has been limited investigation into their properties and no comprehensive comparison between them. In this work, we investigate the loss surfaces from a systematic benchmarking of these approaches across computer vision, natural language processing, and graph learning tasks. The results lead to a simple hypothesis: since both approaches find different flat solutions, combining them should improve generalization even further. We verify this improves over either flat-minima approach in 39 out of 42 cases. When it does not, we investigate potential reasons. We hope our results across image, graph, and text data will help researchers to improve deep learning optimizers, and practitioners to pinpoint the optimizer for the problem at hand.