In this work, we consider the problem of sequence-to-sequence alignment for signals containing outliers. Assuming the absence of outliers, the standard Dynamic Time Warping (DTW) algorithm efficiently computes the optimal alignment between two (generally) variable-length sequences. While DTW is robust to temporal shifts and dilations of the signal, it fails to align sequences in a meaningful way in the presence of outliers that can be arbitrarily interspersed in the sequences. To address this problem, we introduce Drop-DTW, a novel algorithm that aligns the common signal between the sequences while automatically dropping the outlier elements from the matching. The entire procedure is implemented as a single dynamic program that is efficient and fully differentiable. In our experiments, we show that Drop-DTW is a robust similarity measure for sequence retrieval and demonstrate its effectiveness as a training loss on diverse applications. With Drop-DTW, we address temporal step localization on instructional videos, representation learning from noisy videos, and cross-modal representation learning for audio-visual retrieval and localization. In all applications, we take a weakly-or unsupervised approach and demonstrate state-of-the-art results under these settings.

We focus on the task of learning a single index model $\sigma(w^\star \cdot x)$ with respect to the isotropic Gaussian distribution in $d$ dimensions. Prior work has shown that the sample complexity of learning $w^\star$ is governed by the information exponent $k^\star$ of the link function $\sigma$, which is defined as the index of the first nonzero Hermite coefficient of $\sigma$. Ben Arous et al. (2021) showed that $n \gtrsim d^{k^\star-1}$ samples suffice for learning $w^\star$ and that this is tight for online SGD. However, the CSQ lower bound for gradient based methods only shows that $n \gtrsim d^{k^\star/2}$ samples are necessary. In this work, we close the gap between the upper and lower bounds by showing that online SGD on a smoothed loss learns $w^\star$ with $n \gtrsim d^{k^\star/2}$ samples. We also draw connections to statistical analyses of tensor PCA and to the implicit regularization effects of minibatch SGD on empirical losses.

A new era of computing just got closer, as researchers have created the design and run the first ever practical test for an artificial synapse that could let computers replicate some of the brain's most powerful and intricate functions. While computers might seem more powerful than our brains, we can actually deal with a much wider range of possible signals than the "on" and "off" of binary, thanks to the synapses that handle the connections between neurons. Replicating that capability in a computer requires artificial synapses that can reliably send all those subtly different signals. As they describe in Monday's issue of the journal Nature Materials, researchers at the Massachusetts Institute of Technology have performed what they call the first ever practical test of such an artificial synapse, unleashing what's known as neuromorphic computing. While the tests only happened in computer simulations, the tests were promising.

Perhaps you heard recently about a new algorithm that can drive a car? Or scan a picture and find your face in a crowd? It seems as though every week companies are finding new uses for algorithms that adapt as they encounter new data. Last year Wired quoted an ex-Google employee as saying that "Everything in the company is really driven by machine learning." Machine learning has tremendous potential to transform companies, but in practice it's mostly far more mundane than robot drivers and chefs.

Since it launched in 2009, the Kepler Space Telescope has done a good job of collecting data -- too good for human analysis alone. The telescope has produced 14 billion data points about 200,000 stars. It has also amassed 35,000 signals indicating possible planets. People alone would not have been able to keep up. Since it launched in 2009, the Kepler Space Telescope has done a good job of collecting data -- too good for human analysis alone.

Perhaps you heard recently about a new algorithm that can drive a car? Or scan a picture and find your face in a crowd? It seems as though every week companies are finding new uses for algorithms that adapt as they encounter new data. Last year Wired quoted an ex-Google employee as saying that "Everything in the company is really driven by machine learning." Machine learning has tremendous potential to transform companies, but in practice it's mostly far more mundane than robot drivers and chefs.

Philip Swarm Images and Understanding: Thoughts about Images, Ideas about Understanding, H. Barlow, C. Blakemore, and M. Weston-Smith, eds., A collection of essays based on a Rank Prize Fund's International Symposium, organized with the help of Jonathan Miller and held at the Royal Society in October 1986, Cambridge University Press, Cambridge, United Kingdom, 1990, 401 pp., ISBN O-521-34177-9 (cloth), ISBN O-521-36944-4 (paper). This volume is a well-written, informative, and thought-provoking collection of essays that should interest anyone concerned with the psychology of vision and visual communication. The aim of the original symposium was to bring together people from the arts and sciences who could present different perspectives on the subject of images and understanding. The result is an informal tour conducted by leading specialists (predominantly British) that visits both famous scientific battlefields and quaint artistic backwaters. Numerous striking pictures enliven the book: Here you can find the sensory somatic cortex of a bat, the British miners' leader Arthur Scargill in full rant, a notation for ballet, a mole used to advertise British Gas, instructions for righting a caravan, and many others.

On 15-16 November 1989, I attended the Massachusetts Institute of Technology (MIT) Industrial Liaison Program entitled "Networks and Learning." The topic was neural networks, their power, potential, and promise. A dozen distinguished professors and researchers presented informative and entertaining talks to an audience of technically minded business executives and industrial researchers who subscribe to MIT's popular series of symposia offered through their Industrial Liaison Program. The Massachusetts Institute of Technology (MIT) Industrial Liaison Program, "Networks and Learning," was held 15-16 November 1989 at MIT. A dozen distinguished professors and researchers presented informative and entertaining talks to an audience of technically minded business executives and industrial researchers who subscribe to MIT's popular series of symposia offered through its Industrial Liaison Program.

The t.estbed simulates a class of a distributed knowledge-based THERE ARE TWO MAJOR T IEMES of this article. First, WC introduce readers to the emerging subdiscipline of AI called Dzstrrbuted Problem Solving, and more specifically the authors' research on Functionally Accurate, Cooperative systems Second, we discuss the st,ructure of tools that allow more thorough experimentation than has typically been performed in AI research An examplr of such a tool, the Distributed Vehicle Monitoring Testbed, will bc presented. The testbed simulates a class of dist,ributed knowledge-based problem solving systems operating on an abstracted version of a vehicle monitoring task. This presentation emphasizes how the t,estbed is structured to facilit,ate the study of a wide range of issues faced in t,he design of distributed problem solving networks. Distribut,ed Problem Solving (also called Distributed Al) combines the research interests of the fields of AI and Distributed Processing (Chandrasekaran 1981; Davis 1980, 1982; Fehling & Erman 1983).

Philip Swarm Images and Understanding: Thoughts about Images, Ideas about Understanding, H. Barlow, C. Blakemore, and M. Weston-Smith, eds., A collection of essays based on a Rank Prize Fund's International Symposium, organized with the help of Jonathan Miller and held at the Royal Society in October 1986, Cambridge University Press, Cambridge, United Kingdom, 1990, 401 pp., ISBN O-521-34177-9 (cloth), ISBN O-521-36944-4 (paper). This volume is a well-written, informative, and thought-provoking collection of essays that should interest anyone concerned with the psychology of vision and visual communication. The aim of the original symposium was to bring together people from the arts and sciences who could present different perspectives on the subject of images and understanding. The result is an informal tour conducted by leading specialists (predominantly British) that visits both famous scientific battlefields and quaint artistic backwaters. Numerous striking pictures enliven the book: Here you can find the sensory somatic cortex of a bat, the British miners' leader Arthur Scargill in full rant, a notation for ballet, a mole used to advertise British Gas, instructions for righting a caravan, and many others.