The strong lottery ticket hypothesis has highlighted the potential for training deep neural networks by pruning, which has inspired interesting practical and theoretical insights into how neural networks can represent functions. For networks with ReLU activation functions, it has been proven that a target network with depth L can be approximated by the subnetwork of a randomly initialized neural network that has double the target's depth 2L and is wider by a logarithmic factor. We show that a depth L+1 is sufficient. This result indicates that we can expect to find lottery tickets at realistic, commonly used depths while only requiring logarithmic overparametrization. Our novel construction approach applies to a large class of activation functions and is not limited to ReLUs. Code is available on Github (RelationalML/LT-existence).

There have been long-standing controversies and inconsistencies over the experiment setup and criteria for identifying the winning ticket in literature. To reconcile such, we revisit the definition of lottery ticket hypothesis, with comprehensive and more rigorous conditions. Under our new definition, we show concrete evidence to clarify whether the winning ticket exists across the major DNN architectures and/or applications. Through extensive experiments, we perform quantitative analysis on the correlations between winning tickets and various experimental factors, and empirically study the patterns of our observations. We find that the key training hyperparameters, such as learning rate and training epochs, as well as the architecture characteristics such as capacities and residual connections, are all highly correlated with whether and when the winning tickets can be identified. Based on our analysis, we summarize a guideline for parameter settings in regards of specific architecture characteristics, which we hope to catalyze the research progress on the topic of lottery ticket hypothesis.

We consider the problem of building a reinforcement learning (RL) agent that can both accomplish non-trivial tasks, like winning a real-time strategy game, and strictly follow high-level language commands from humans, like "attack", even if a command is sub-optimal. We call this novel yet important problem, Grounded Reinforcement Learning (GRL). Compared with other language grounding tasks, GRL is particularly non-trivial and cannot be simply solved by pure RL or behavior cloning (BC). From the RL perspective, it is extremely challenging to derive a precise reward function for human preferences since the commands are abstract and the valid behaviors are highly complicated and multi-modal. From the BC perspective, it is impossible to obtain perfect demonstrations since human strategies in complex games are typically sub-optimal. We tackle GRL via a simple, tractable, and practical constrained RL objective and develop an iterative RL algorithm, REinforced demonstration Distillation (RED), to obtain a strong GRL policy. We evaluate the policies derived by RED, BC and pure RL methods on a simplified real-time strategy game, MiniRTS. Experiment results and human studies show that the RED policy is able to consistently follow human commands and achieve a higher win rate than the baselines. We release our code and present more examples at https://sites.google.com/view/grounded-rl.

We investigate the power of voting among diverse, randomized software agents. With teams of computer Go agents in mind, we develop a novel theoretical model of two-stage noisy voting that builds on recent work in machine learning. This model allows us to reason about a collection of agents with different biases (determined by the first-stage noise models), which, furthermore, apply randomized algorithms to evaluate alternatives and produce votes (captured by the second-stage noise models). We analytically demonstrate that a uniform team, consisting of multiple instances of any single agent, must make a significant number of mistakes, whereas a diverse team converges to perfection as the number of agents grows. Our experiments, which pit teams of computer Go agents against strong agents, provide evidence for the effectiveness of voting when agents are diverse.

With the constant increase of available data in various domains, such as the Internet of Things, Social Networks or Smart Cities, it has become fundamental that agents are able to process and reason with such data in real time. Whereas reasoning over time-annotated data with background knowledge may be challenging, due to the volume and velocity in which such data is being produced, such complex reasoning is necessary in scenarios where agents need to discover potential problems and this cannot be done with simple stream processing techniques. Stream Reasoners aim at bridging this gap between reasoning and stream processing and LASER is such a stream reasoner designed to analyse and perform complex reasoning over streams of data. It is based on LARS, a rule-based logical language extending Answer Set Programming, and it has shown better runtime results than other state-of-the-art stream reasoning systems. Nevertheless, for high levels of data throughput even LASER may be unable to compute answers in a timely fashion. In this paper, we study whether Convolutional and Recurrent Neural Networks, which have shown to be particularly well-suited for time series forecasting and classification, can be trained to approximate reasoning with LASER, so that agents can benefit from their high processing speed.

Three years ago, on a drab, chilly summer day in the Dutch port of Den Helder, Amazon made an extravagant pitch for its first-ever big-budget video game, Breakaway. The event, streamed live on Twitch, was an esports tournament with a twist: It would take place on a 355-foot-long naval patrol ship, the kind that hunts down pirates and drug smugglers in the Caribbean. On the upper decks, sailors stood taut as the camera ogled the vessel's 76mm cannon. Then a pair of emcees from Amazon Game Studios, occasionally shouting over the thrum of passing helicopters, introduced the competitors. They were down below, huddled around high-end monitors--headsets on, knees jiggling anxiously, cans of Red Bull cracked open.

A successful intelligent control of patient food for treatment purpose must combines patient interesting food list and doctors efficient treatment food list. Actually, many rural communities in Sudan have extremely limited access to diabetic diet centers. People travel long distances to clinics or medical facilities, and there is a shortage of medical experts in most of these facilities. This results in slow service, and patients end up waiting long hours without receiving any attention. Hence diabetic diet expert systems can play a significant role in such cases where medical experts are not readily available. This paper presents the design and implementation of an intelligent medical expert system for diabetes diet that intended to be used in Sudan. The development of the proposed expert system went through a number of stages such problem and need identification, requirements analysis, knowledge acquisition, formalization, design and implementation. Visual prolog was used for designing the graphical user interface and the implementation of the system. The proposed expert system is a promising helpful tool that reduces the workload for physicians and provides diabetics with simple and valuable assistance.

In response to this need, Shelby Lyman, the host of past Public Broadcasting Station (PBS) series on world chess championship matches, organized the AGS Challenge Match at the New School for Social Research in New York City. Funding for this event was provided by AGS Computers, Inc., a New Jersey-based software firm. The match was held September 22-25, with one game played each day, and was widely covered by the international press. Participating were Hitech, at 2407 then the highest-rated computer in the world, and International Grandmaster Arnold S. Denker, a former U.S. champion. Denker's rating of 2410 was comparable to that of Hitech.

Hitech wins the Pennsylvania State Chess Championship. In a field of 46 players, including 8 masters, Carnegie-Mellon University's Hitech won the title outright with a score of 4.5 - 0.5. The finale was very exciting as the number one seeded player, International Master Ed Formanek, had a perfect 4-0 going into the final round against the number two seed Hitech, which was the only competitor with 3.5 - 0.5. Hitech had to win to gain the title, whereas Formanek could become Champion by drawing this game. Hitech played beautifully, and the game culminated in a brilliant endgame that Hitech handled perfectly.

To the list of French accomplishments you may now add "robot basketball training" -- at least if the video above is to be believed. But you probably shouldn't believe it when members of Poitiers Basket 86 testify that amusement park rides improved the team's "spatial orientation" and helped them defeat top-ranked Chalon. It'd be different if the "robots" were teaching them perfect free-throw or helping them walk, obviously, but PB86 is known for its innovative advertising, and this seems like a quirky example. Hit the video above to see the pranksters at work, but know that, as with Sartre and Camus, something gets lost in translation.