The central building block of convolutional neural networks (CNNs) is the convolution operator, which enables networks to construct informative features by fusing both spatial and channel-wise information within local receptive fields at each layer. A broad range of prior research has investigated the spatial component of this relationship, seeking to strengthen the representational power of a CNN by enhancing the quality of spatial encodings throughout its feature hierarchy. In this work, we focus instead on the channel relationship and propose a novel architectural unit, which we term the “Squeeze-and-Excitation” (SE) block, that adaptively recalibrates channel-wise feature responses by explicitly modelling interdependencies between channels. We show that these blocks can be stacked together to form SENet architectures that generalise extremely effectively across different datasets. We further demonstrate that SE blocks bring significant improvements in performance for existing state-of-the-art CNNs at slight additional computational cost. Squeeze-and-Excitation Networks formed the foundation of our ILSVRC 2017 classification submission which won first place and reduced the top-5 error to 2.251 percent, surpassing the winning entry of 2016 by a relative improvement of ${\sim }$ ∼ 25 percent. Models and code are available at https://github.com/hujie-frank/SENet .

/pdf/squeeze-and-excitation-networks-50icgwqgeu.pdf

Squeeze-and-Excitation Networks

https://www2.econ.iastate.edu/tesfatsi/DeepLearningInNeuralNetworksOverview.JSchmidhuber2015.pdf

Deep learning in neural networks

Machine Learning is the study of methods for programming computers to learn. Computers are applied to a wide range of tasks, and for most of these it is relatively easy for programmers to design and implement the necessary software. However, there are many tasks for which this is difficult or impossible. These can be divided into four general categories. First, there are problems for which there exist no human experts. For example, in modern automated manufacturing facilities, there is a need to predict machine failures before they occur by analyzing sensor readings. Because the machines are new, there are no human experts who can be interviewed by a programmer to provide the knowledge necessary to build a computer system. A machine learning system can study recorded data and subsequent machine failures and learn prediction rules. Second, there are problems where human experts exist, but where they are unable to explain their expertise. This is the case in many perceptual tasks, such as speech recognition, hand-writing recognition, and natural language understanding. Virtually all humans exhibit expert-level abilities on these tasks, but none of them can describe the detailed steps that they follow as they perform them. Fortunately, humans can provide machines with examples of the inputs and correct outputs for these tasks, so machine learning algorithms can learn to map the inputs to the outputs. Third, there are problems where phenomena are changing rapidly. In finance, for example, people would like to predict the future behavior of the stock market, of consumer purchases, or of exchange rates. These behaviors change frequently, so that even if a programmer could construct a good predictive computer program, it would need to be rewritten frequently. A learning program can relieve the programmer of this burden by constantly modifying and tuning a set of learned prediction rules. Fourth, there are applications that need to be customized for each computer user separately. Consider, for example, a program to filter unwanted electronic mail messages. Different users will need different filters. It is unreasonable to expect each user to program his or her own rules, and it is infeasible to provide every user with a software engineer to keep the rules up-to-date. A machine learning system can learn which mail messages the user rejects and maintain the filtering rules automatically. Machine learning addresses many of the same research questions as the fields of statistics, data mining, and psychology, but with differences of emphasis. Statistics focuses on understanding the phenomena that have generated the data, often with the goal of testing different hypotheses about those phenomena. Data mining seeks to find patterns in the data that are understandable by people. Psychological studies of human learning aspire to understand the mechanisms underlying the various learning behaviors exhibited by people (concept learning, skill acquisition, strategy change, etc.).

Machine learning

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

Data Mining Practical Machine Learning Tools and Techniques

In procedural content generation, one is often interested in generating a large number of artifacts that are not only of high quality but also diverse, in terms of gameplay, visual impression or some other criterion. We investigate several search-based approaches to creating good and diverse game content, in particular approaches based on evolution strategies with or without diversity preservation mechanisms, novelty search and random search. The content domain is game levels, more precisely map sketches for strategy games, which are meant to be used as suggestions in the Sentient Sketchbook design tool. Several diversity metrics are possible for this type of content: we investigate tile-based, objective-based and visual impression distance. We find that evolution with diversity preservation mechanisms can produce both good and diverse content, but only when using appropriate distance measures. Reversely, we can draw conclusions about the suitability of these distance measures for the domain from the comparison of diversity preserving versus blind restart evolutionary algorithms.

/pdf/searching-for-good-and-diverse-game-levels-4h4cp1udur.pdf

Searching for good and diverse game levels

This paper describes the MetaCompose music generator,
 a compositional, extensible framework for affective music composition. In this context ‘affective’ refers to the music generator’s ability to express emotional information. The main purpose of MetaCompose is to create music in real-time that can express different mood-states, which we achieve through a unique combination of a graph traversal-based chord sequence generator, a search-based melody generator, a pattern-based accompaniment generator, and a theory for mood expression. Melody generation uses a novel evolutionary technique combining FI-2POP with multi-objective optimization. This allows us to explore a Pareto front of diverse solutions that are creatively equivalent under the terms of a multi-criteria objective function. Two quantitative user studies were performed to evaluate the system: one focusing on the music generation technique, and the other that explores valence expression,
 via the introduction of dissonances. The results of these studies demonstrate (i) that each part of the generation system improves the perceived quality of the music produced, and (ii) how valence expression via dissonance produces the perceived affective state. This system, which can reliably generate affect-expressive music, can subsequently be integrated in any kind of interactive application (e.g., games) to create an adaptive and dynamic soundtrack.

Affective evolutionary music composition with MetaCompose

Evolutionary car racing (ECR) is extended to the case of two cars racing on the same track. A sensor representation is devised, and various methods of evolving car controllers for competitive racing are explored. ECR can be combined with co-evolution in a wide variety of ways, and one aspect which is explored here is the relative-absolute fitness continuum. Systematical behavioural differences are found along this continuum; further, a tendency to specialization and the reactive nature of the controller architecture are found to limit evolutionary progress.

/pdf/arms-races-and-car-races-1qwhzr65lp.pdf

Arms races and car races

While Monte Carlo tree search (MCTS) methods have shown promise in a variety of different board games, more complex video games still present significant challenges. Recently, several modifications to the core MCTS algorithm have been proposed with the hope to increase its effectiveness on arcade-style video games. This paper investigates of how well these modifications perform in general video game playing using the general video game AI (GVG-AI) framework and introduces a new MCTS modification called UCT reverse penalty that penalizes the MCTS controller for exploring recently visited children. The results of our experiments show that a combination of two MCTS modifications can improve the performance of the vanilla MCTS controller, but the effectiveness of the modifications highly depends on the particular game being played.

/pdf/investigating-mcts-modifications-in-general-video-game-sl6ccu6l9o.pdf

Investigating MCTS modifications in general video game playing

Deep reinforcement learning (RL) has shown impressive results in a variety of domains, learning directly from high-dimensional sensory streams. However, when neural networks are trained in a fixed environment, such as a single level in a video game, they will usually overfit and fail to generalize to new levels. When RL models overfit, even slight modifications to the environment can result in poor agent performance. This paper explores how procedurally generated levels during training can increase generality. We show that for some games procedural level generation enables generalization to new levels within the same distribution. Additionally, it is possible to achieve better performance with less data by manipulating the difficulty of the levels in response to the performance of the agent. The generality of the learned behaviors is also evaluated on a set of human-designed levels. The results suggest that the ability to generalize to human-designed levels highly depends on the design of the level generators. We apply dimensionality reduction and clustering techniques to visualize the generators' distributions of levels and analyze to what degree they can produce levels similar to those designed by a human.

Julian Togelius

Papers

Searching for good and diverse game levels

Affective evolutionary music composition with MetaCompose

Arms races and car races

Investigating MCTS modifications in general video game playing

Procedural Level Generation Improves Generality of Deep Reinforcement Learning