CS 510: Lecture 8. Deep Learning, Fairness, and Bias

Transcription

1 CS 510: Lecture 8 Deep Learning, Fairness, and Bias

2 Next Week All Presentations, all the time Upload your presentation before class if using slides Sign up for a timeslot google doc, if you haven t already done so

3 Artificial Neural Networks: History Belief that it was necessary to model underlying brain architecture for AI In contrast to encoded symbolic knowledge (best represented by expert systems) Hebb - learning is altering strength of synaptic connections

4 Neural Networks Attempt to build a computation system based on the parallel architecture of brains Characteristics: Many simple processing elements Many connections Simple messages Adaptive interaction

5 Brains neurons of > 20 types, synapses, 1ms 10ms cycle time Signals are noisy spike trains of electrical potential Axonal arborization Synapse Axon from another cell Dendrite Axon Nucleus Synapses Cell body or Soma Chapter 20, Section 5 3

6 Benefits of NN User friendly (well, reasonably) Non-linear Noise tolerant Many applications Credit fraud/assignment Robotic Control

7 Neurons Inputs (either from outside or other neurons ) Weighted connections that correspond to synaptic efficiency Threshold values to weight the inputs Passed through activation function to determine output

8 Example Unit Binary input/output Rule 1 if w0*i0 + w1*i1 +wb > 0 0 if w0*i0 + w1*i1 +wb <= 0

9 Activation functions g(in i ) g(in i ) (a) in i (b) in i (a) is a step function or threshold function (b) is a sigmoid function 1/(1 + e x ) Changing the bias weight W 0,i moves the threshold location Note similarity to logistic regression... Chapter 20, Section 5 5

10 -0.06 W1-2.5 W2 f(x) W3 1.4

11 f(x) x = =

12 How to Adapt? Perceptron Learning Rule change the weight by an amount proportional to the difference between the desired output and the actual output. As an equation: ΔWi = η * (D - Y)Ii, where D is desired output and Y is actual output Stop when converges

13 Limits of Perceptrons Minsky and Papert 1969 Fails on linearly inseparable instances XOR linearly separable - pattern space can be separated by single hyperplane

14 Perceptrons vs Decision Trees

15 Multilayer Perceptrons (MLP)

16 Back Propagation Start with a set of known examples (supervised approach) Assign random initial weights Run examples through and calculate the mean-squared error Propagate the error by making small changes to the weights at each level Use chain rule to calculate the gradient efficiently Lather, rinse, repeat

17 Gradient Descent Algorithm Have some function Want Outline: Start with some Keep changing to reduce until we hopefully end up at a minimum

18 The gradient of J ( J) at a point can be thought of as a vector indicating which way is uphill J(θ 0,θ 1 ) If J is an error function we want to move downhill - opposite to the gradient

19 Gradient descent algorithm Have function J Want to produce vectors s.t. J(θ1)>J(θ2)>... start w/ θ0 θi+1 = θi - ɑi J(θi) ɑ(alpha) is the learning rate

20 Stochastic Gradient Descent Update J every time you look at a training example

21 Some non-linear activation functions

22 Most common activation function

23 A dataset Fields class etc

24 Training the neural network Fields class etc

25 Training data Fields class etc Initialise with random weights

26 Training data Fields class etc Present a training pattern

27 Training data Fields class etc 1.4 Feed it through to get output

28 Training data Fields class etc 1.4 Compare with target output error 0.8

29 Training data Fields class etc 1.4 Adjust weights based on error error 0.8

30 Training data Fields class etc Present a training pattern

31 Training data Fields class etc 6.4 Feed it through to get output

32 Training data Fields class etc 6.4 Compare with target output error -0.1

33 Training data Fields class etc 6.4 Adjust weights based on error error -0.1

34 Training data Fields class etc 6.4 And so on error -0.1 Repeat this thousands, maybe millions of times each time taking a random training instance, and making slight weight adjustments Algorithms for weight adjustment are designed to make changes that will reduce the error

35 The decision boundary perspective Initial random weights

36 The decision boundary perspective Present a training instance / adjust the weights

37 The decision boundary perspective Present a training instance / adjust the weights

38 The decision boundary perspective Present a training instance / adjust the weights

39 The decision boundary perspective Present a training instance / adjust the weights

40 The decision boundary perspective Eventually.

41 The point I am trying to make weight-learning algorithms for NNs are dumb they work by making thousands and thousands of tiny adjustments, each making the network do better at the most recent pattern, but perhaps a little worse on many others but, by dumb luck, eventually this tends to be good enough to learn effective classifiers for many real applications

42 Some other points Detail of a standard NN weight learning algorithm later If f(x) is non-linear, a network with 1 hidden layer can, in theory, learn perfectly any classification problem. A set of weights exists that can produce the targets from the inputs. The problem is finding them.

43 Some other by the way points If f(x) is linear, the NN can only draw straight decision boundaries (even if there are many layers of units)

44 Some other by the way points NNs use nonlinear f(x) so they can draw complex boundaries, but keep the data unchanged

45 Some other by the way points NNs use nonlinear f(x) so they can draw complex boundaries, but keep the data unchanged SVMs only draw straight lines, but they transform the data first in a way that makes that OK

46 Neural network vocabulary Neuron = logistic regression or similar function Input layer = input training/test vector Bias unit = intercept term/always on feature Activation = response Activation function is a logistic (or similar sigmoid nonlinearity) Backpropagation = running stochastic gradient descent across a multilayer network Weight decay - regularization or Bayesian prior

47 Deep Learning Most current machine learning works well because of human-designed representations and input features Machine learning becomes just optimizing weights to best make a final prediction Representation learning attempts to automatically learn good features or representations Deep learning algorithms attempt to learn multiple levels of representation of increasing complexity/abstraction

48 Deep Architecture

49 Deep Learning Overview Train networks with many layers (vs. shallow nets with just a couple of layers) Multiple layers work to build an improved feature space First layer learns 1st order features (e.g. edges ) 2nd layer learns higher order features (combinations of first layer features, combinations of edges, etc.) In current models layers often learn in an unsupervised mode and discover general features of the input space serving multiple tasks related to the unsupervised instances (image recognition, etc.) Then final layer features are fed into supervised layer(s) And entire network is often subsequently tuned using supervised training of the entire net, using the initial weightings learned in the unsupervised phase Could also do fully supervised versions, etc. (early BP attempts)

50 Why Deep Learning?

51 Learning Representations Handcrafting features is time-consuming incomplete domain/... The features are often both over-specified and The work has to be done again for each task/ We must move beyond handcrafted features and simple ML Humans develop representations for learning and reasoning Our computers should do the same

52 The Curse of Dimensionality

53 Unsupervised Feature and Weight Learning Today, most practical, good NLP& ML methods require labeled training data (i.e., supervised learning) But almost all data is unlabeled Most information must be acquired unsupervised Fortunately, a good model of observed data can really help you learn classification decisions

54 Learning Multiple Levels of Representation

55 Successive Layers Learn Deeper Representations object models object parts (combination of edges) edges pixels

56 Impressive Results Especially on Large Datasets Object Recognition - better than anything out there Speech Recognition (google voice search) Many other perceptual tasks in vision and NLP

57 Why now? Bigger Data - deep learning works best Better Hardware - multicore CPUs and GPUs Better Algorithms - autoencoders, deep belief networks, etc Let us train multiple inner layers well

58 Breakthrough: Unsupervised Pre-training

59 Difficulties with Supervised Networks Early layers of MLP do not get trained well Diffusion of Gradient error attenuates as it propagates to earlier layers Leads to very slow training Exacerbated since top couple layers can usually learn any task "pretty well" and thus the error to earlier layers drops quickly as the top layers "mostly" solve the task lower layers never get the opportunity to use their capacity to improve results, they just do a random feature map Need a way for early layers to do effective work Often not enough labeled data available while lots of unlabeled data Can we use unsupervised/semi-supervised approaches to take advantage of the unlabeled data Deep networks tend to have more local minima problems than shallow networks during supervised training

60 Semi-supervised Learning

61 Semi-supervised Learning

62 Training Deep Networks Build a feature space Note that this is what we do with SVM kernels, or trained hidden layers in BP, etc., but now we will build the feature space using deep architectures Unsupervised training between layers can decompose the problem into distributed subproblems (with higher levels of abstraction) to be further decomposed at subsequent layers

63 Greedy Layer-wise Training Train first layer using your data without the labels (unsupervised) Since there are no targets at this level, labels don't help. Could also use the more abundant unlabeled data which is not part of the training set (i.e. self-taught learning). Freeze the first layer parameters and start training the second layer using the output of the first layer as the unsupervised input to the second layer Repeat this for as many layers as desired This builds our set of robust features Use the outputs of the final layer as inputs to a supervised layer/model and train the last supervised layer(s) (leave early weights frozen) Unfreeze all weights and fine tune the full network by training with a supervised approach, given the pre-processed weight settings

64 Greedy Layer-wise Training Greedy layer-wise training avoids many of the problems of trying to train a deep net in a supervised fashion Each layer gets full learning focus in its turn since it is the only current "top" layer Can take advantage of the unlabeled data When you finally tune the entire network with supervised training the network weights have already been adjusted so that you are in a good error basin and just need fine tuning. This helps with problems of Ineffective early layer learning Deep network local minima We will discuss the two most common approaches Stacked Auto-Encoders Deep Belief Networks

65 The new way to train multi-layer NNs Train this layer first

66 The new way to train multi-layer NNs Train this layer first then this layer

67 The new way to train multi-layer NNs Train this layer first then this layer then this layer

68 The new way to train multi-layer NNs Train this layer first then this layer then this layer then this layer

69 The new way to train multi-layer NNs Train this layer first then this layer then this layer then this layer finally this layer

70 The new way to train multi-layer NNs EACH of the (non-output) layers is trained to be an auto-encoder Basically, it is forced to learn good features that describe what comes from the previous layer

71 an auto-encoder is trained, with an absolutely standard weight-adjustment algorithm to reproduce the input

72 an auto-encoder is trained, with an absolutely standard weight-adjustment algorithm to reproduce the input By making this happen with (many) fewer units than the inputs, this forces the hidden layer units to become good feature detectors

73 One Auto-encoder 73

74 Stacked Auto-encoders Stack sparse auto-encoders on top of each other, drop decode layer each time 74

75 Stacked auto-encoders Do supervised training on last layer Then do supervised training on whole network to fine tune the weights 75

76 Manifold Learning Hypothesis 76

77 Caveats Prevent the layers from just learning the identity (learn Features instead) Undercomplete - middle layer smaller than input Sparsity - penalize hidden unit activations Use regularization to keep most nodes at or near 0 Denoising - Stochastically corrupt training instance, but train encoder to decode uncorrupted instance Contractive - force encoder to have small derivatives (stay on manifold) 77

78 Fairness and Learning Going to show video of Aylin Link on course website to her talk (may be easier for online students if there is feedback). 78