Deep Learning Authors: Yann LeCun, Yoshua Bengio, Geoffrey Hinton

Slides:



Advertisements
Similar presentations
A brief review of non-neural-network approaches to deep learning
Advertisements

Machine Learning: Connectionist McCulloch-Pitts Neuron Perceptrons Multilayer Networks Support Vector Machines Feedback Networks Hopfield Networks.
Machine Learning Neural Networks
Decision Support Systems
Neural Networks Basic concepts ArchitectureOperation.
Deep Belief Networks for Spam Filtering
Machine Learning Motivation for machine learning How to set up a problem How to design a learner Introduce one class of learners (ANN) –Perceptrons –Feed-forward.
Convolutional Neural Networks for Image Processing with Applications in Mobile Robotics By, Sruthi Moola.
Artificial Neural Nets and AI Connectionism Sub symbolic reasoning.
IE 585 Introduction to Neural Networks. 2 Modeling Continuum Unarticulated Wisdom Articulated Qualitative Models Theoretic (First Principles) Models Empirical.
 The most intelligent device - “Human Brain”.  The machine that revolutionized the whole world – “computer”.  Inefficiencies of the computer has lead.
NEURAL NETWORKS FOR DATA MINING
LeCun, Bengio, And Hinton doi: /nature14539
 Based on observed functioning of human brain.  (Artificial Neural Networks (ANN)  Our view of neural networks is very simplistic.  We view a neural.
Deep Convolutional Nets
Introduction to Neural Networks Introduction to Neural Networks Applied to OCR and Speech Recognition An actual neuron A crude model of a neuron Computational.
Neural Networks Teacher: Elena Marchiori R4.47 Assistant: Kees Jong S2.22
Object Recognizing. Deep Learning Success in 2012 DeepNet and speech processing.
Business Intelligence and Decision Support Systems (9 th Ed., Prentice Hall) Chapter 6: Artificial Neural Networks for Data Mining.
Business Intelligence and Decision Support Systems (9 th Ed., Prentice Hall) Chapter 6: Artificial Neural Networks for Data Mining.
Neural networks (2) Reminder Avoiding overfitting Deep neural network Brief summary of supervised learning methods.
語音訊號處理之初步實驗 NTU Speech Lab 指導教授: 李琳山 助教: 熊信寬
Deep Learning Overview Sources: workshop-tutorial-final.pdf
Lecture 12. Outline of Rule-Based Classification 1. Overview of ANN 2. Basic Feedforward ANN 3. Linear Perceptron Algorithm 4. Nonlinear and Multilayer.
Machine Learning Artificial Neural Networks MPλ ∀ Stergiou Theodoros 1.
A Presentation on Adaptive Neuro-Fuzzy Inference System using Particle Swarm Optimization and it’s Application By Sumanta Kundu (En.R.No.
Xintao Wu University of Arkansas Introduction to Deep Learning 1.
Business Intelligence and Decision Support Systems (9 th Ed., Prentice Hall) Chapter 6: Artificial Neural Networks for Data Mining.
Machine Learning Supervised Learning Classification and Regression
2/13/2018 4:38 AM © Microsoft Corporation. All rights reserved. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN.
Big data classification using neural network
Convolutional Neural Network
CS 388: Natural Language Processing: LSTM Recurrent Neural Networks
The Relationship between Deep Learning and Brain Function
Deep Learning Amin Sobhani.
Randomness in Neural Networks
Data Mining, Neural Network and Genetic Programming
Data Mining, Neural Network and Genetic Programming
Recurrent Neural Networks for Natural Language Processing
COMP24111: Machine Learning and Optimisation
Matt Gormley Lecture 16 October 24, 2016
ICS 491 Big Data Analytics Fall 2017 Deep Learning
Intelligent Information System Lab
Different Units Ramakrishna Vedantam.
Neural networks (3) Regularization Autoencoder
Supervised Training of Deep Networks
Deep learning and applications to Natural language processing
Deep Belief Networks Psychology 209 February 22, 2013.
Dynamic Routing Using Inter Capsule Routing Protocol Between Capsules
CSE P573 Applications of Artificial Intelligence Neural Networks
Introduction to Deep Learning for neuronal data analyses
Grid Long Short-Term Memory
Neural Networks Advantages Criticism
Image Captions With Deep Learning Yulia Kogan & Ron Shiff
Recurrent Neural Networks
Recurrent Neural Networks
Deep Architectures for Artificial Intelligence
CSE 573 Introduction to Artificial Intelligence Neural Networks
network of simple neuron-like computing elements
The Big Health Data–Intelligent Machine Paradox
Long Short Term Memory within Recurrent Neural Networks
Neural Networks Geoff Hulten.
Artificial Intelligence 10. Neural Networks
实习生汇报 ——北邮 张安迪.
Neural networks (3) Regularization Autoencoder
Presentation By: Eryk Helenowski PURE Mentor: Vincent Bindschaedler
Automatic Handwriting Generation
Ch4: Backpropagation (BP)
Bidirectional LSTM-CRF Models for Sequence Tagging
Presentation transcript:

Deep Learning Authors: Yann LeCun, Yoshua Bengio, Geoffrey Hinton Presented by Nick Harris

Outline The Success of Deep Learning The Basics of Neural Networks Forward and Back propagation Supervised Learning Convolutional Neural Networks Recurrent Neural Networks The Future of Deep Learning

Success of Deep Learning Deep Learning has become increasingly popular in recent years It is increasingly used in consumer products such as smartphones, and in commercial websites It has also greatly improved state-of-the-art performance in a wide variety of tasks such as speech and object recognition What makes Deep Learning so successful?

Success (continued) Deep Learning offers numerous advantages over older conventional machine learning techniques: Deep Learning techniques have a greater capacity to process raw data Conventional techniques often require a lot of domain expertise and precise engineering for good pattern recognition. NNs on the other hand can perform well with little prior knowledge. Deep Learning techniques tend to continue to improve when given vastly larger amounts of data Two Deep Learning models that have been especially successful are the Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN)

Neural Networks Neural Network systems consist of large networks of artificial neurons These neurons map a set of inputs X to an output y Parameters: weights (w), bias (b) Activation Function (phi) to introduce nonlinearity

Neural Networks (continued) Some popular activation functions:

Neural Networks (continued) Many neurons are combined to form a network. Maps an arbitrary number of inputs X to and arbitrary number of outputs Y A Neural Network with even one hidden layer is sufficient to compute any function Deep Learning is where many successive layers are used. These many layers learn increasingly abstract representation of the data Learn much better than small network in practice

Forward/Back Propagation Forward Propagation is the process of activating the network with to produce an output Back Propagation is the process of learning by adjusting the weights of the network

Gradient Descent Learning can be thought of as finding an optimal point in a high dimensional landscape Computing the error gradient and adjusting the weights can be thought of as “hill-climbing” where we move progressively toward the optimal point Stochastic Gradient Descent is a technique used in practice that speeds up learning by computing the average gradient over a random set of samples (but introduces noise).

Supervised Learning One of the most common learning processes for neural networks Asks the network to make a prediction, then use the correct answer to adjust the weights with back propagation Requires large data sets

Convolutional Neural Nets (CNNs) A special class of neural networks that excel at processing images Uses Convolutional and Pooling layers to break down images

CNNs (continued) Convolutional Layer Pooling Layer

Value of CNNs CNNs have proven to be extremely good at image classification and object detection. As a result they are widely used in the computer vision field.

Recurrent Neural Networks (RNNs) RNNs are a class of neural networks that excel at processing sequential data (sensitive to time and order) One especially successful application is natural language processing This includes predicting the next character of text, or the next word in a sequence A more sophisticated application is translation by mapping text to a “thought vector”, then mapping to text in another language

RNNs (continued) RNNs process input in a sequence one step at a time, receiving inputs from previous time steps Maintain a hidden “state vector” that contains information about the history of past elements in the sequence RNNs were initially problematic to train, but recent advances in architecture and training have allowed them to produce state-of-the-art results Can be augmented with a more explicit memory to form an Long Short Term Memory (LSTM) network, which has proven to be more powerful

Future of Deep Learning Unsupervised Learning is not as popular as Supervised of late, but is considered to have great potential Interest in combining CNNs and RNNs with reinforcement learning for difficult computer vision tasks Interest in improving natural language processing performance

Questions?