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

Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources. To address this limitation, we introduce "deep compression", a three stage pipeline: pruning, trained quantization and Huffman coding, that work together to reduce the storage requirement of neural networks by 35x to 49x without affecting their accuracy. Our method first prunes the network by learning only the important connections. Next, we quantize the weights to enforce weight sharing, finally, we apply Huffman coding. After the first two steps we retrain the network to fine tune the remaining connections and the quantized centroids. Pruning, reduces the number of connections by 9x to 13x; Quantization then reduces the number of bits that represent each connection from 32 to 5. On the ImageNet dataset, our method reduced the storage required by AlexNet by 35x, from 240MB to 6.9MB, without loss of accuracy. Our method reduced the size of VGG-16 by 49x from 552MB to 11.3MB, again with no loss of accuracy. This allows fitting the model into on-chip SRAM cache rather than off-chip DRAM memory. Our compression method also facilitates the use of complex neural networks in mobile applications where application size and download bandwidth are constrained. Benchmarked on CPU, GPU and mobile GPU, compressed network has 3x to 4x layerwise speedup and 3x to 7x better energy efficiency.

https://forum.stanford.edu/events/posterslides/DeepCompressionCompressingDeepNeuralNetworkswithPruning,TrainedQuantizationandHuffmanCoding.pdf

Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding

Planning algorithms are impacting technical disciplines and industries around the world, including robotics, computer-aided design, manufacturing, computer graphics, aerospace applications, drug design, and protein folding. This coherent and comprehensive book unifies material from several sources, including robotics, control theory, artificial intelligence, and algorithms. The treatment is centered on robot motion planning but integrates material on planning in discrete spaces. A major part of the book is devoted to planning under uncertainty, including decision theory, Markov decision processes, and information spaces, which are the “configuration spaces” of all sensor-based planning problems. The last part of the book delves into planning under differential constraints that arise when automating the motions of virtually any mechanical system. Developed from courses taught by the author, the book is intended for students, engineers, and researchers in robotics, artificial intelligence, and control theory as well as computer graphics, algorithms, and computational biology.

Planning Algorithms: Introductory Material

Model Checking

A classic introduction to artificial intelligence intended to bridge the gap between theory and practice, "Principles of Artificial Intelligence" describes fundamental AI ideas that underlie applications such as natural language processing, automatic programming, robotics, machine vision, automatic theorem proving, and intelligent data retrieval. Rather than focusing on the subject matter of the applications, the book is organized around general computational concepts involving the kinds of data structures used, the types of operations performed on the data structures, and the properties of the control strategies used. "Principles of Artificial Intelligence"evolved from the author's courses and seminars at Stanford University and University of Massachusetts, Amherst, and is suitable for text use in a senior or graduate AI course, or for individual study.

Principles of Artificial Intelligence

1. INTRODUCTION 1.1 Background and Motivations A distributed system is a network G=(V,E) of |V|=n processors connected by |E|=e direct communication links, where each processor has a local non-shared memory and can communicate by sending messages to and receiving messages from its neighbours. The behaviour of these processors can be conveniently described as finite-state and event-driven; that is, each processor at any time is in a particular state and, when a predefined event occurs (e.g., a message is received, the internal clock reaches a predetermined value, etc.), it will serially perform some operations whose nature depends on the current state and on the occurred event. The operations that can be performed are local computations, transmission of messages, and change of state. A distributed algorithm is the specification of what operations must be serially performed by a processor when a predefined event occurs in a given state. To ensure a fully distributed computation in the system, it is assumed that every processor has the same algorithm. A fundamental computation in this environment is the election process; that is, the process of changing from an initial system configuration, where every processor in the network is in the same state, to a final configuration where exactly one processor is in a predefined state (say, elected) and all other processors are in another predefined state (say, defeated). Note that there is no a-priori restriction on which processor should become elected.

/pdf/guessing-games-and-distributed-computations-in-synchronous-iide0foq8l.pdf

Guessing Games and Distributed Computations in Synchronous Networks

We develop a new perspective on trees, that enables us to distinguish and analyse many different subclasses of known classes of (height-)balanced search trees in a uniform manner. The approach shows that a great many different local constraints, including an arbitrary degree of density, can be enforced on everyday balanced search tree models, without losing the O(log n) bound on the time for insertions, deletions and finds. The theory extends known concepts from the study of B-trees.

/pdf/stratified-balanced-search-trees-3o3kayfdru.pdf

Stratified balanced search trees

The information processing capabilities of artificial living (AL) systems are far more powerful than commonly believed. Modelling single organisms by means of so-called cognitive transducers, we estimate the computational power of AL systems by viewing them as conglomerates of organisms. We describe a scenario in which an AL system is engaged in a potentially unbounded, unpredictable interaction with an environment, to which it can react by learning and adjusting its behaviour. By means of lineages of cognitive transducers we also model the evolution of AL systems. Among the examples are "communities of agents", i.e., communities of mobile, interactive cognitive transducers. Most AL systems show the emergence of a computational power that is not present at the level of the individual organisms. Indeed, in all but trivial cases the resulting systems possess a super-Turing computing power. This means that the systems cannot be simulated by traditional computational models like Turing machines and may in principle solve non-computable tasks. The results derive from recent results from the theory of interactive evolutionary computing systems in terms of AL systems.

/pdf/the-emergent-computational-potential-of-evolving-artificial-3z7bjaypoh.pdf

The emergent computational potential of evolving artificial living systems

The power of certain heuristic rules is indicated by the relative reduction in the complexity of computations carried out, due to the use of the heuristics. A concept of complexity is needed to evaluate the performance of programs as they operate with a varying set of heuristic rules in use. We present such a complexity measure, which we have found useful for the quantitative evaluation of strategies in the context of a long-term research project on poker. We define our measure in terms of the level-complexity for decision trees and study the measure for the relevant class of decision trees with a fixed but arbitrary number of levels, h, and k leaves (all at the last level). We determine the best and worst case distributions for the levels in this measure. We show, for instance, that for the smallest value, Lh(k), and the largest value, Uh(k), which the level-complexity for such trees can have, limk→∞ Lh(k)/logh(k) = 1 and limk→∞ Uh(k)/log(k) = 1. We show also that the level-entropy assumes its maximum value of log(k) just when the path-entropy reaches its minimum value over all trees with k leaves but, in general, the values of either measure can vary in a rather unrelated manner. We give a detailed example of how the complexity measure is used in evaluating the power of heuristic rules used in assessing the performance of the Quasi-Optimizer (QO), a program currently under development. The objectives of the QO are explained, and the three main phases of operation of the program are described within the framework of the poker project.

On the Complexity of Decision Trees, the Quasi-Optimizer, and the Power of Heuristic Rules

For a number of common configurations of points (lines) in the plane, we develop datastructures in which insertions and deletions of points (or lines, respectively) can be processed rapidly without sacrificing much of the efficiency of query answering of known static structures for these configurations. As a main result we establish a fully dynamic maintenance algorithm for convex hulls that can process insertions and deletions of single points in only O(log3n) steps or less per transaction, where n is the number of points currently in the set. The algorithm has several intriguing applications, including that one can “peel” a set of n points in only O(log3n) steps and that one can maintain two sets at a costs of only O(log3n) or less per insertion and deletion such that it never takes more than O(log2n) steps to determine whether the two sets are separable by a straight line. Also efficient algorithms are obtained for dynamically maintaining the common intersection of a set of half-spaces and for dynamically maintaining the maximal elements of a set of plane points. The results are all derived by means of one master technique, which is applied repeatedly and which seems to capture an appropriate notion of “decomposability” for configurations.

Jan van Leeuwen

Papers

Guessing Games and Distributed Computations in Synchronous Networks

Stratified balanced search trees

The emergent computational potential of evolving artificial living systems

On the Complexity of Decision Trees, the Quasi-Optimizer, and the Power of Heuristic Rules

Dynamically maintaining configurations in the plane (Detailed Abstract)