Neural Networks for Machine Learning
Lecture 11a
Hopfield Nets
Geoffrey Hinton
Nitish Srivastava,
Kevin Swersky
Tijmen Tieleman
Abdel-rahman Mohamed
Hopfield Nets
A Hopfield net is composed of
binary
In Neural Computation, 3, pages 79-87.
Adaptive Mixtures of Local Experts
Robert A. Jacobs
Michael I. Jordan
Department of Brain & Cognitive Sciences
Massachusetts Institute of Technology
Cambridge, M
Learning Dynamical System
Models from Data
CS 294-112: Deep Reinforcement Learning
Week 3, Lecture 1
Sergey Levine
Overview
1. Before: learning to act by imitating a human
2. Last lecture: choose good
Markov Decision Processes
and Solving Finite Problems
February 8, 2017
Overview of Upcoming Lectures
Feb 8:
Feb 13:
Feb 15:
Feb 22:
Markov decision processes, value iteration, policy iteration
Policy
Q-Function Learning Methods
February 15, 2017
Value Functions
I
Definitions (review):
Q (s, a) = E r0 + r1 + 2 r2 + . . . | s0 = s, a0 = a
Called Q-function or state-action-value function
V (s) = E r0
Learning Policies by Imitating
Optimal Control
CS 294-112: Deep Reinforcement Learning
Week 3, Lecture 2
Sergey Levine
Overview
1. Last time: learning models of system dynamics and using optimal
contr
Policy Gradient Methods
February 13, 2017
Policy Optimization Problems
maximize E [expression]
I
I
I
I
I
Fixed-horizon episodic:
PT 1
t=0 rt
P
T 1
Average-cost: limT T1 t=0
rt
P t
Infinite-horizon dis
Optimal Control, Trajectory
Optimization, and Planning
CS 294-112: Deep Reinforcement Learning
Week 2, Lecture 2
Sergey Levine
Announcements
1. Assignment 1 will be out next week
2. Friday section
Rev
Beyond Learning from Reward
reward
Mnih et al. 15
reinforcement learning agent
what is the reward?
In the real world, humans dont get a score.
video from Montessori New Zealand
Tesauro 95
Kohl & Stone
Convolutional networks
Sebastian Seung
Convolution (1D)
(w s) i = w i j s j = w j si j
j
j
Convolutional network
Neural network with spatial organization
every neuron has a location
usually on a gr
ECE-340
Spring 2008
Probabilistic Methods in Engineering (3 credits)
M, W 3:00-4:15 PM
Room: Dane Smith Hall 325
Syllabus
Course Goals: To introduce the student to basic theoretical concepts and compu
Pooling
Sebastian Seung
Translation-invariant
recognition
DNF network
The images of a single object
are highly variable.
Illumination
Scale
Translation
Rotation
Deformation
Objects in the same class a
Contemporary ConvNets
Sebastian Seung
ImageNet Large-Scale Visual
Recognition Challenge
LSVRC-2010
subset of ImageNet
1.2 million training images
1000 x 1000 categories
50,000 validation images
Training neural nets
Sebastian Seung
Two challenges
How to make training error lower?
Gradient learning is finicky and must be
How to improve generalization to inputs
not seen in training?
Next le