We consider linear prediction with a convex Lipschitz loss, or more generally, stochastic convex optimization problems of generalized linear form, i.e.~where each instantaneous loss is a scalar convex function of a linear function. We show that in this setting, early stopped Gradient Descent (GD), without any explicit regularization or projection, ensures excess error at most $\varepsilon$ (compared to the best possible with unit Euclidean norm) with an optimal, up to logarithmic factors, sample complexity of $\tilde{O}(1/\varepsilon^2)$ and only $\tilde{O}(1/\varepsilon^2)$ iterations. This contrasts with general stochastic convex optimization, where $\Omega(1/\varepsilon^4)$ iterations are needed Amir et al. 2021. The lower iteration complexity is ensured by leveraging uniform convergence rather than stability. But instead of uniform convergence in a norm ball, which we show can guarantee suboptimal learning using $\Theta(1/\varepsilon^4)$ samples, we rely on uniform convergence in a distribution-dependent ball.

We propose a novel method for computing exact pointwise robustness of deep neural networks for all convex lp norms. Our algorithm, GeoCert, finds the largest lp ball centered at an input point x0, within which the output class of a given neural network with ReLU nonlinearities remains unchanged. We relate the problem of computing pointwise robustness of these networks to that of computing the maximum norm ball with a fixed center that can be contained in a non-convex polytope. This is a challenging problem in general, however we show that there exists an efficient algorithm to compute this for polyhedral complices. Further we show that piecewise linear neural networks partition the input space into a polyhedral complex.

DeepMind's attempt to teach an AI to play soccer started with a virtual player writhing around on the floor--so it nailed at least one aspect of the game right from kickoff. But pinning down the mechanics of the beautiful game--from basics like running and kicking to higher-order concepts like teamwork and tackling--proved a lot more challenging, as new research from the Alphabet-backed AI firm demonstrates. The work--published this week in the journal Science Robotics--might seem frivolous, but learning the fundamentals of soccer could one day help robots to move around our world in more natural, more human ways. "In order to'solve' soccer, you have to actually solve lots of open problems on the path to artificial general intelligence [AGI]," says Guy Lever, a research scientist at DeepMind. "There's controlling the full humanoid body, coordination--which is really tough for AGI--and actually mastering both low-level motor control and things like long-term planning."

Finding the perfect holiday gift can be maddening (is this the color they'd want? Is it something they already have? Is it so last year?), but really, once you have a sense of a person's taste, it's not impossible. This season, we'll be talking to members of various tribes to find out exactly what to get that college student, or serious home cook, or boss (who has everything) in your life. Think of it as a window into their brain trust--or, at least, a very helpful starting point.

Robotic soccer is an ideal task to demonstrate new techniques and explore new problems. Moreover, problems and solutions can easily be communicated because soccer is a well-known game. Our intention in building a robotic soccer team and participating in RoboCup-98 was, first, to demonstrate the usefulness of the self-localization methods we have developed. Second, we wanted to show that playing soccer based on an explicit world model is much more effective than other methods. Third, we intended to explore the problem of building and maintaining a global team world model.

JEEVES's success depended crucially on JEEVES's visual range was extremely JEEVES as successful as it was? JEEVES's success was that its software JEEVES's hardware was designed and built by JEEVES can reverse the direction of the brush. It is equipped with seven ultrasonic proximity sensors (only five were used in the competition), a wide-angle color camera, and a high-speed colorbased vision system manufactured by Newton Research Labs. Prior to the competition, the vision system was trained to recognize yellow tennis balls, pink squiggle balls, and cyan markers that marked the gate. The vision system proved extremely reliable during the competition, benefiting from clear color cues provided by the objects.

Picture tactile feedback and situated computing. That's the year when a revolving cadre of scientists began work on the problem of predicting the outcome of the spin of a roulette wheel. Although lacking the societal import of, say, predicting cancer in a patient, or even poison in a mushroom, predicting roulette seems on the face of it of even greater difficulty. The game itself is designed in every way for unpredictability. The problem is at its core a machine learning problem with a direct physical basis.

This work deals with designing simple behaviors to allow quadruped robots to play soccer. The robots are fully autonomous; they cannot exchange messages between each other. They are equipped with a charge-coupled-device camera that allows them to detect objects in the scene. In addition to vision problems such as changing lighting conditions and color confusion, legged robots must cope with "bouncing images" because of successive legs hitting the ground. When defining task-driven strategies, the designer has to take into account the influences of the locomotion and vision systems on the behavior.

Individual agent skills, such as kicking and dribbling (running with the ball), are important prerequisites for team collaboration. For each of these skills, many parameters affect the details of the skill execution. For example, in the ball skill of dribbling, there are parameters that affect how quickly the agent runs, how far ahead it kicks the ball, and on which side of its body the agent keeps the ball while it dribbles. The settings for these parameters usually involve a tradeoff, such as speed versus safety or power versus accuracy. It is important to gain an understanding of what exactly these tradeoffs are before "correct" parameter settings can be made. We created a trainer client that connects to the server as an omniscient offline coach client.

Sony has provided a robot platform for research and development in physical agents, namely, fully autonomous legged robots. In this article, we describe our work using Sony's legged robots to participate at the RoboCup-98 legged robot demonstration and competition. Robotic soccer represents a challenging environment for research in systems with multiple robots that need to achieve concrete objectives, particularly in the presence of an adversary. Furthermore, RoboCup offers an excellent opportunity for robot entertainment. We introduce the RoboCup context and briefly present Sony's legged robot.