DOCSLIB.ORG

Backpropagation and the Brain the Structure That Is Only Implicit in the Raw Sensory Input

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Backpropagation and the Brain the Structure That Is Only Implicit in the Raw Sensory Input PERSPECTIVES predominantly unsupervised fashion1,25–27, building representations that make explicit Backpropagation and the brain the structure that is only implicit in the raw sensory input. It is natural to wonder, then, Timothy P. Lillicrap , Adam Santoro, Luke Marris, Colin J. Akerman and whether backprop has anything to tell us Geoffrey Hinton about learning in the brain25,28–30. Here we argue that in spite of these Abstract | During learning, the brain modifies synapses to improve behaviour. In the apparent differences, the brain has the capacity cortex, synapses are embedded within multilayered networks, making it difficult to implement the core principles underlying to determine the effect of an individual synaptic modification on the behaviour of backprop. The main idea is that the brain the system. The backpropagation algorithm solves this problem in deep artificial could compute effective synaptic updates by using feedback connections to induce neuron neural networks, but historically it has been viewed as biologically problematic. activities whose locally computed differences Nonetheless, recent developments in neuroscience and the successes of artificial encode backpropagation-like error signals. neural networks have reinvigorated interest in whether backpropagation offers We link together a seemingly disparate set insights for understanding learning in the cortex. The backpropagation algorithm of learning algorithms into this framework, learns quickly by computing synaptic updates using feedback connections to which we call ‘neural gradient representation 9,27,31–41 deliver error signals. Although feedback connections are ubiquitous in the cortex, by activity differences’ (NGRAD) . The NGRAD framework demonstrates that it is difficult to see how they could deliver the error signals required by strict it is possible to embrace the core principles formulations of backpropagation. Here we build on past and recent developments of backpropagation while sidestepping to argue that feedback connections may instead induce neural activities whose many of its problematic implementation differences can be used to locally approximate these signals and hence drive requirements. These considerations may be effective learning in deep networks in the brain. relevant to any brain circuit that incorporates both feedforward and feedback connectivity. We nevertheless focus on the cortex, which The brain learns by modifying the synaptic that compute synaptic changes that reduce is defined by its multilaminar structure connections between neurons1–5. Although the error (Fig. 1). and hierarchical organization, and so has synaptic physiology helps explain the In machine learning, backpropagation of long been viewed as exhibiting many of rules and processes behind individual error (‘backprop’)7–10 is the algorithm most the architectural features associated with modifications, it does not explain how often used to train deep neural networks deep networks. individual modifications coordinate to (BOx 1) and is the most successful learning achieve a network’s goal. Since learning procedure for these networks. Networks Credit assignment in networks cannot be just a blind accumulation of trained with backprop are at the heart This article emphasizes the role of learning myopic, synapse-specific events that do of recent successes of machine learning, in the generation of adaptive behaviour. not consider downstream behavioural including state-of-the-art speech11 and It should be acknowledged that brains consequences, we need to uncover the image recognition12,13, as well as language undoubtedly have prior knowledge that has principles orchestrating plasticity across translation14. Backprop also underpins recent been optimized by evolution (that is, in the whole networks if we are to understand progress in unsupervised learning problems form of neural architectures and default learning in the brain. such as image and speech generation15,16, connectivity strengths). Priors may ensure Within machine learning, researchers language modelling17 and other next-step that only a limited amount of learning based study ways of coordinating synaptic updates prediction tasks18. In addition, combining on a relatively small amount of task error or to improve performance in artificial neural backprop with reinforcement learning has feedback information is needed throughout networks, without being constrained by given rise to significant advances in solving an animal’s lifetime to acquire all the biological reality. They start by defining control problems, such as mastering Atari skills the animal will exhibit. Nonetheless, the architecture of the neural network, games19 and beating top human professionals although animals often display impressive which comprises the number of neurons in the games of Go20,21 and poker22. behaviours from birth, they are also capable and how they are connected. For example, Backprop uses error signals that are of extraordinary feats that could not have investigators often use deep networks sent through feedback connections to been tuned by evolution but instead require with many layers of neurons, since these adjust synapses and has classically been long bouts of learning. Some examples of architectures have proved to be very effective described in the supervised learning setting such feats in humans are playing Go and for many tasks. Next, researchers define (that is, with explicit, externally provided chess; programming a computer or designing an error function6 that quantifies how poorly targets). However, the brain appears to a video game; writing and playing a piano the network is currently achieving its goals use its feedback connections for different concerto; learning the vocabularies and and then they search for learning algorithms purposes23,24 and is thought to learn in a grammars of multiple languages; recognizing NATURE REVIEWS | NEUROSCIENCE PERSPECTIVES a No feedback Scalar feedback Vector feedback Synapse undergoing learning Feedback signal (e.g. gradient) Feedforward Hebbian Perturbation Backpropagation Backprop-like learning Feedback neuron (required for learning) network learning learning with feedback network Feedforward neuron (required for learning) Output Diffuse scalar reinforcement signal c Error landscape Perturbation learning Input b Scalar feedback Vector feedback 1 Parameter Weight perturbation Backpropagation Hebbian learning Precision of synaptic change in reducing error Backpropagation Node perturbation Backpropagation approximations Parameter 2 Fig. 1 | A spectrum of learning algorithms. a | Left to right: a neural b | Backpropagation and perturbation algorithms fall along a spectrum with network computes an output through a series of simple computational respect to the specificity of the synaptic change they prescribe. c | Algorithms units. To improve its outputs for a task , it adjusts the synapses between on this spectrum learn at different speeds. Without feedback , synaptic these units. Simple Hebbian learning — which dictates that a synaptic con- parameters wander randomly on the error surface. Scalar feedback does not nection should strengthen if a presynaptic neuron reliably contributes to a require detailed feedback circuits, but it learns slowly. Since the same signal postsynaptic neuron’s firing — cannot make meaningful changes to the blue is used to inform learning at all synapses, the difficulty of deciding whether synapse, because it does not consider this synapse’s downstream effect on to strengthen or weaken a synapse scales with the number of synapses in the the network output. Perturbation methods measure the change in error network: if millions of synapses are changed simultaneously , the effect of one caused by random perturbations to neural activities (node perturbation) or synapse change is swamped by the noise created by all the other changes, synapse strengths44 (weight perturbation) and use this measured change as and it takes millions of trials to average away this noise43–46. The inverse scal- a global scalar reinforcement signal that controls whether a proposed ing of learning speed with network size makes global reinforcement methods perturbation is accepted or rejected. The backprop algorithm instead extremely slow , even for moderately sized neural networks. Precise vector computes the synapse update required in order to most quickly reduce the feedback via backprop learns quickly. In real networks, it is not possible to error. In backprop, vector error signals are delivered backward along make perfect use of the internal structure of the network to compute per- the original path of influence for a neuron. In the brain, vector feedback synapse changes, but the brain may have discovered ways to approximate might be delivered in a variety of ways, including via a separate network. the speed of backprop. thousands of objects; and diagnosing a influence — its projective field — rapidly of a network is captured by an error medical problem and performing vascular expands, so the effect of changing the function that computes the degree to microsurgery. Recent work in machine synapse strength depends on the strengths which the network’s outputs [y1,…,yM] learning suggests that these behaviours of many subsequent synapses in the network deviate from their target values [t1,…,tM] depend on powerful and general learning (for example, the blue connections spreading — for example, via the squared error, algorithms12,20. Our interest here, then, is in from the input layer in
Load more

Recommended publications
  • Backpropagation and Deep Learning in the Brain
    Backpropagation and Deep Learning in the Brain Simons Institute -- Computational Theories of the Brain 2018 Timothy Lillicrap DeepMind, UCL With: Sergey Bartunov, Adam Santoro, Jordan Guerguiev, Blake Richards, Luke Marris, Daniel Cownden, Colin Akerman, Douglas Tweed, Geoffrey Hinton The “credit assignment” problem The solution in artificial networks: backprop Credit assignment by backprop works well in practice and shows up in virtually all of the state-of-the-art supervised, unsupervised, and reinforcement learning algorithms. Why Isn’t Backprop “Biologically Plausible”? Why Isn’t Backprop “Biologically Plausible”? Neuroscience Evidence for Backprop in the Brain? A spectrum of credit assignment algorithms: A spectrum of credit assignment algorithms: A spectrum of credit assignment algorithms: How to convince a neuroscientist that the cortex is learning via [something like] backprop - To convince a machine learning researcher, an appeal to variance in gradient estimates might be enough. - But this is rarely enough to convince a neuroscientist. - So what lines of argument help? How to convince a neuroscientist that the cortex is learning via [something like] backprop - What do I mean by “something like backprop”?: - That learning is achieved across multiple layers by sending information from neurons closer to the output back to “earlier” layers to help compute their synaptic updates. How to convince a neuroscientist that the cortex is learning via [something like] backprop 1. Feedback connections in cortex are ubiquitous and modify the
    [Show full text]
  • Training Autoencoders by Alternating Minimization
    Under review as a conference paper at ICLR 2018 TRAINING AUTOENCODERS BY ALTERNATING MINI- MIZATION Anonymous authors Paper under double-blind review ABSTRACT We present DANTE, a novel method for training neural networks, in particular autoencoders, using the alternating minimization principle. DANTE provides a distinct perspective in lieu of traditional gradient-based backpropagation techniques commonly used to train deep networks. It utilizes an adaptation of quasi-convex optimization techniques to cast autoencoder training as a bi-quasi-convex optimiza- tion problem. We show that for autoencoder configurations with both differentiable (e.g. sigmoid) and non-differentiable (e.g. ReLU) activation functions, we can perform the alternations very effectively. DANTE effortlessly extends to networks with multiple hidden layers and varying network configurations. In experiments on standard datasets, autoencoders trained using the proposed method were found to be very promising and competitive to traditional backpropagation techniques, both in terms of quality of solution, as well as training speed. 1 INTRODUCTION For much of the recent march of deep learning, gradient-based backpropagation methods, e.g. Stochastic Gradient Descent (SGD) and its variants, have been the mainstay of practitioners. The use of these methods, especially on vast amounts of data, has led to unprecedented progress in several areas of artificial intelligence. On one hand, the intense focus on these techniques has led to an intimate understanding of hardware requirements and code optimizations needed to execute these routines on large datasets in a scalable manner. Today, myriad off-the-shelf and highly optimized packages exist that can churn reasonably large datasets on GPU architectures with relatively mild human involvement and little bootstrap effort.
    [Show full text]
  • Q-Learning in Continuous State and Action Spaces
    -Learning in Continuous Q State and Action Spaces Chris Gaskett, David Wettergreen, and Alexander Zelinsky Robotic Systems Laboratory Department of Systems Engineering Research School of Information Sciences and Engineering The Australian National University Canberra, ACT 0200 Australia [cg dsw alex]@syseng.anu.edu.au j j Abstract. -learning can be used to learn a control policy that max- imises a scalarQ reward through interaction with the environment. - learning is commonly applied to problems with discrete states and ac-Q tions. We describe a method suitable for control tasks which require con- tinuous actions, in response to continuous states. The system consists of a neural network coupled with a novel interpolator. Simulation results are presented for a non-holonomic control task. Advantage Learning, a variation of -learning, is shown enhance learning speed and reliability for this task.Q 1 Introduction Reinforcement learning systems learn by trial-and-error which actions are most valuable in which situations (states) [1]. Feedback is provided in the form of a scalar reward signal which may be delayed. The reward signal is defined in relation to the task to be achieved; reward is given when the system is successfully achieving the task. The value is updated incrementally with experience and is defined as a discounted sum of expected future reward. The learning systems choice of actions in response to states is called its policy. Reinforcement learning lies between the extremes of supervised learning, where the policy is taught by an expert, and unsupervised learning, where no feedback is given and the task is to find structure in data.
    [Show full text]
  • The Deep Learning Revolution and Its Implications for Computer Architecture and Chip Design
    The Deep Learning Revolution and Its Implications for Computer Architecture and Chip Design Jeffrey Dean Google Research [email protected] Abstract The past decade has seen a remarkable series of advances in machine learning, and in particular deep learning approaches based on artificial neural networks, to improve our abilities to build more accurate systems across a broad range of areas, including computer vision, speech recognition, language translation, and natural language understanding tasks. This paper is a companion paper to a keynote talk at the 2020 International Solid-State Circuits Conference (ISSCC) discussing some of the advances in machine learning, and their implications on the kinds of computational devices we need to build, especially in the post-Moore’s Law-era. It also discusses some of the ways that machine learning may also be able to help with some aspects of the circuit design process. Finally, it provides a sketch of at least one interesting direction towards much larger-scale multi-task models that are sparsely activated and employ much more dynamic, example- and task-based routing than the machine learning models of today. Introduction The past decade has seen a remarkable series of advances in machine learning (ML), and in particular deep learning approaches based on artificial neural networks, to improve our abilities to build more accurate systems across a broad range of areas [LeCun et al. 2015]. Major areas of significant advances ​ ​ include computer vision [Krizhevsky et al. 2012, Szegedy et al. 2015, He et al. 2016, Real et al. 2017, Tan ​ ​ ​ ​ ​ ​ and Le 2019], speech recognition [Hinton et al.
    [Show full text]
  • Double Backpropagation for Training Autoencoders Against Adversarial Attack
    1 Double Backpropagation for Training Autoencoders against Adversarial Attack Chengjin Sun, Sizhe Chen, and Xiaolin Huang, Senior Member, IEEE Abstract—Deep learning, as widely known, is vulnerable to adversarial samples. This paper focuses on the adversarial attack on autoencoders. Safety of the autoencoders (AEs) is important because they are widely used as a compression scheme for data storage and transmission, however, the current autoencoders are easily attacked, i.e., one can slightly modify an input but has totally different codes. The vulnerability is rooted the sensitivity of the autoencoders and to enhance the robustness, we propose to adopt double backpropagation (DBP) to secure autoencoder such as VAE and DRAW. We restrict the gradient from the reconstruction image to the original one so that the autoencoder is not sensitive to trivial perturbation produced by the adversarial attack. After smoothing the gradient by DBP, we further smooth the label by Gaussian Mixture Model (GMM), aiming for accurate and robust classification. We demonstrate in MNIST, CelebA, SVHN that our method leads to a robust autoencoder resistant to attack and a robust classifier able for image transition and immune to adversarial attack if combined with GMM. Index Terms—double backpropagation, autoencoder, network robustness, GMM. F 1 INTRODUCTION N the past few years, deep neural networks have been feature [9], [10], [11], [12], [13], or network structure [3], [14], I greatly developed and successfully used in a vast of fields, [15]. such as pattern recognition, intelligent robots, automatic Adversarial attack and its defense are revolving around a control, medicine [1]. Despite the great success, researchers small ∆x and a big resulting difference between f(x + ∆x) have found the vulnerability of deep neural networks to and f(x).
    [Show full text]
  • ARCHITECTS of INTELLIGENCE for Xiaoxiao, Elaine, Colin, and Tristan ARCHITECTS of INTELLIGENCE
    MARTIN FORD ARCHITECTS OF INTELLIGENCE For Xiaoxiao, Elaine, Colin, and Tristan ARCHITECTS OF INTELLIGENCE THE TRUTH ABOUT AI FROM THE PEOPLE BUILDING IT MARTIN FORD ARCHITECTS OF INTELLIGENCE Copyright © 2018 Packt Publishing All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, without the prior written permission of the publisher, except in the case of brief quotations embedded in critical articles or reviews. Every effort has been made in the preparation of this book to ensure the accuracy of the information presented. However, the information contained in this book is sold without warranty, either express or implied. Neither the author, nor Packt Publishing or its dealers and distributors, will be held liable for any damages caused or alleged to have been caused directly or indirectly by this book. Packt Publishing has endeavored to provide trademark information about all of the companies and products mentioned in this book by the appropriate use of capitals. However, Packt Publishing cannot guarantee the accuracy of this information. Acquisition Editors: Ben Renow-Clarke Project Editor: Radhika Atitkar Content Development Editor: Alex Sorrentino Proofreader: Safis Editing Presentation Designer: Sandip Tadge Cover Designer: Clare Bowyer Production Editor: Amit Ramadas Marketing Manager: Rajveer Samra Editorial Director: Dominic Shakeshaft First published: November 2018 Production reference: 2201118 Published by Packt Publishing Ltd. Livery Place 35 Livery Street Birmingham B3 2PB, UK ISBN 978-1-78913-151-2 www.packt.com Contents Introduction ........................................................................ 1 A Brief Introduction to the Vocabulary of Artificial Intelligence .......10 How AI Systems Learn ........................................................11 Yoshua Bengio .....................................................................17 Stuart J.
    [Show full text]
  • Unsupervised Speech Representation Learning Using Wavenet Autoencoders Jan Chorowski, Ron J
    1 Unsupervised speech representation learning using WaveNet autoencoders Jan Chorowski, Ron J. Weiss, Samy Bengio, Aaron¨ van den Oord Abstract—We consider the task of unsupervised extraction speaker gender and identity, from phonetic content, properties of meaningful latent representations of speech by applying which are consistent with internal representations learned autoencoding neural networks to speech waveforms. The goal by speech recognizers [13], [14]. Such representations are is to learn a representation able to capture high level semantic content from the signal, e.g. phoneme identities, while being desired in several tasks, such as low resource automatic speech invariant to confounding low level details in the signal such as recognition (ASR), where only a small amount of labeled the underlying pitch contour or background noise. Since the training data is available. In such scenario, limited amounts learned representation is tuned to contain only phonetic content, of data may be sufficient to learn an acoustic model on the we resort to using a high capacity WaveNet decoder to infer representation discovered without supervision, but insufficient information discarded by the encoder from previous samples. Moreover, the behavior of autoencoder models depends on the to learn the acoustic model and a data representation in a fully kind of constraint that is applied to the latent representation. supervised manner [15], [16]. We compare three variants: a simple dimensionality reduction We focus on representations learned with autoencoders bottleneck, a Gaussian Variational Autoencoder (VAE), and a applied to raw waveforms and spectrogram features and discrete Vector Quantized VAE (VQ-VAE). We analyze the quality investigate the quality of learned representations on LibriSpeech of learned representations in terms of speaker independence, the ability to predict phonetic content, and the ability to accurately re- [17].
    [Show full text]
  • Approaching Hanabi with Q-Learning and Evolutionary Algorithm
    St. Cloud State University theRepository at St. Cloud State Culminating Projects in Computer Science and Department of Computer Science and Information Technology Information Technology 12-2020 Approaching Hanabi with Q-Learning and Evolutionary Algorithm Joseph Palmersten [email protected] Follow this and additional works at: https://repository.stcloudstate.edu/csit_etds Part of the Computer Sciences Commons Recommended Citation Palmersten, Joseph, "Approaching Hanabi with Q-Learning and Evolutionary Algorithm" (2020). Culminating Projects in Computer Science and Information Technology. 34. https://repository.stcloudstate.edu/csit_etds/34 This Starred Paper is brought to you for free and open access by the Department of Computer Science and Information Technology at theRepository at St. Cloud State. It has been accepted for inclusion in Culminating Projects in Computer Science and Information Technology by an authorized administrator of theRepository at St. Cloud State. For more information, please contact [email protected]. Approaching Hanabi with Q-Learning and Evolutionary Algorithm by Joseph A Palmersten A Starred Paper Submitted to the Graduate Faculty of St. Cloud State University In Partial Fulfillment of the Requirements for the Degree of Master of Science in Computer Science December, 2020 Starred Paper Committee: Bryant Julstrom, Chairperson Donald Hamnes Jie Meichsner 2 Abstract Hanabi is a cooperative card game with hidden information that requires cooperation and communication between the players. For a machine learning agent to be successful at the Hanabi, it will have to learn how to communicate and infer information from the communication of other players. To approach the problem of Hanabi the machine learning methods of Q- learning and Evolutionary algorithm are proposed as potential solutions.
    [Show full text]
  • Linear Prediction-Based Wavenet Speech Synthesis
    LP-WaveNet: Linear Prediction-based WaveNet Speech Synthesis Min-Jae Hwang Frank Soong Eunwoo Song Search Solution Microsoft Naver Corporation Seongnam, South Korea Beijing, China Seongnam, South Korea [email protected] [email protected] [email protected] Xi Wang Hyeonjoo Kang Hong-Goo Kang Microsoft Yonsei University Yonsei University Beijing, China Seoul, South Korea Seoul, South Korea [email protected] [email protected] [email protected] Abstract—We propose a linear prediction (LP)-based wave- than the speech signal, the training and generation processes form generation method via WaveNet vocoding framework. A become more efficient. WaveNet-based neural vocoder has significantly improved the However, the synthesized speech is likely to be unnatural quality of parametric text-to-speech (TTS) systems. However, it is challenging to effectively train the neural vocoder when the target when the prediction errors in estimating the excitation are database contains massive amount of acoustical information propagated through the LP synthesis process. As the effect such as prosody, style or expressiveness. As a solution, the of LP synthesis is not considered in the training process, the approaches that only generate the vocal source component by synthesis output is vulnerable to the variation of LP synthesis a neural vocoder have been proposed. However, they tend to filter. generate synthetic noise because the vocal source component is independently handled without considering the entire speech To alleviate this problem, we propose an LP-WaveNet, production process; where it is inevitable to come up with a which enables to jointly train the complicated interactions mismatch between vocal source and vocal tract filter.
    [Show full text]
  • Fast Neural Network Emulation of Dynamical Systems for Computer Animation
    Fast Neural Network Emulation of Dynamical Systems for Computer Animation Radek Grzeszczuk 1 Demetri Terzopoulos 2 Geoffrey Hinton 2 1 Intel Corporation 2 University of Toronto Microcomputer Research Lab Department of Computer Science 2200 Mission College Blvd. 10 King's College Road Santa Clara, CA 95052, USA Toronto, ON M5S 3H5, Canada Abstract Computer animation through the numerical simulation of physics-based graphics models offers unsurpassed realism, but it can be computation­ ally demanding. This paper demonstrates the possibility of replacing the numerical simulation of nontrivial dynamic models with a dramatically more efficient "NeuroAnimator" that exploits neural networks. Neu­ roAnimators are automatically trained off-line to emulate physical dy­ namics through the observation of physics-based models in action. De­ pending on the model, its neural network emulator can yield physically realistic animation one or two orders of magnitude faster than conven­ tional numerical simulation. We demonstrate NeuroAnimators for a va­ riety of physics-based models. 1 Introduction Animation based on physical principles has been an influential trend in computer graphics for over a decade (see, e.g., [1, 2, 3]). This is not only due to the unsurpassed realism that physics-based techniques offer. In conjunction with suitable control and constraint mechanisms, physical models also facilitate the production of copious quantities of real­ istic animation in a highly automated fashion. Physics-based animation techniques are beginning to find their way into high-end commercial systems. However, a well-known drawback has retarded their broader penetration--compared to geometric models, physical models typically entail formidable numerical simulation costs. This paper proposes a new approach to creating physically realistic animation that differs Emulation for Animation 883 radically from the conventional approach of numerically simulating the equations of mo­ tion of physics-based models.
    [Show full text]
  • Natural Reweighted Wake-Sleep
    Natural Reweighted Wake-Sleep Csongor Várady Riccardo Volpi Max Planck Institute for Romanian Institute of Mathematics in the Sciences Science and Technology Leipzig, Germany Cluj-Napoca, Romania [email protected] [email protected] Luigi Malagò Nihat Ay Romanian Institute of Max Planck Institute for Science and Technology Mathematics in the Sciences Cluj-Napoca, Romania Leipzig, Germany [email protected] [email protected] Abstract Natural gradient has been successfully employed in a wide range of optimization problems. However, for the training of neural networks the resulting increase in computational complexity sets a limitation to its practical application. Helmholtz Machines are a particular type of generative models, composed of two Sigmoid Belief Networks, commonly trained using the Wake-Sleep algorithm. The locality of the connections in this type of networks induces sparsity and a particular structure for the Fisher information matrix that can be exploited for the evaluation of its inverse, allowing the efficient computation of the natural gradient also for large networks. We introduce a novel algorithm called Natural Reweighted Wake-Sleep, a geometric adaptation of Reweighted Wake-Sleep, based on the computation of the natural gradient. We present an experimental analysis of the algorithm in terms of speed of convergence and the value of the log-likelihood, both with respect to number of iterations and training time, demonstrating improvements over non-geometric baselines. 1 Introduction Deep generative models have been successfully employed in unsupervised learning to model complex and high dimensional distributions thanks to their ability to extract higher-order representations of the data and thus generalize better [16, 8].
    [Show full text]
  • Neural Networks for Machine Learning Lecture 4A Learning To
    Neural Networks for Machine Learning Lecture 4a Learning to predict the next word Geoffrey Hinton with Nitish Srivastava Kevin Swersky A simple example of relational information Christopher = Penelope Andrew = Christine Margaret = Arthur Victoria = James Jennifer = Charles Colin Charlotte Roberto = Maria Pierro = Francesca Gina = Emilio Lucia = Marco Angela = Tomaso Alfonso Sophia Another way to express the same information • Make a set of propositions using the 12 relationships: – son, daughter, nephew, niece, father, mother, uncle, aunt – brother, sister, husband, wife • (colin has-father james) • (colin has-mother victoria) • (james has-wife victoria) this follows from the two above • (charlotte has-brother colin) • (victoria has-brother arthur) • (charlotte has-uncle arthur) this follows from the above A relational learning task • Given a large set of triples that come from some family trees, figure out the regularities. – The obvious way to express the regularities is as symbolic rules (x has-mother y) & (y has-husband z) => (x has-father z) • Finding the symbolic rules involves a difficult search through a very large discrete space of possibilities. • Can a neural network capture the same knowledge by searching through a continuous space of weights? The structure of the neural net local encoding of person 2 output distributed encoding of person 2 units that learn to predict features of the output from features of the inputs distributed encoding of person 1 distributed encoding of relationship local encoding of person 1 inputs local encoding of relationship Christopher = Penelope Andrew = Christine Margaret = Arthur Victoria = James Jennifer = Charles Colin Charlotte What the network learns • The six hidden units in the bottleneck connected to the input representation of person 1 learn to represent features of people that are useful for predicting the answer.
    [Show full text]