Gradient Episodic Memory for Continual Learning (2017-06-01T00:00:00.000000Z)

TL;DR

A model for continual learning, called Gradient Episodic Memory (GEM) is proposed that alleviates forgetting, while allowing beneficial transfer of knowledge to previous tasks.

Abstract

One major obstacle towards AI is the poor ability of models to solve new problems quicker, and without forgetting previously acquired knowledge. To better understand this issue, we study the problem of continual learning, where the model observes, once and one by one, examples concerning a sequence of tasks. First, we propose a set of metrics to evaluate models learning over a continuum of data. These metrics characterize models not only by their test accuracy, but also in terms of their ability to transfer knowledge across tasks. Second, we propose a model for continual learning, called Gradient Episodic Memory (GEM) that alleviates forgetting, while allowing beneficial transfer of knowledge to previous tasks. Our experiments on variants of the MNIST and CIFAR-100 datasets demonstrate the strong performance of GEM when compared to the state-of-the-art.

Authors

Marc'Aurelio Ranzato

12 papers

David Lopez-Paz

6 papers

TL;DR

Abstract

Authors

References54 items

Statistical Learning Theory

Encoder Based Lifelong Learning

Training Mixture Models at Scale via Coresets

Improved multitask learning through synaptic intelligence

CommAI: Evaluating the first steps towards a useful general AI

PathNet: Evolution Channels Gradient Descent in Super Neural Networks

Overcoming catastrophic forgetting in neural networks

iCaRL: Incremental Classifier and Representation Learning

Expert Gate: Lifelong Learning with a Network of Experts

Less-forgetting Learning in Deep Neural Networks

Learning without Forgetting

Learning feed-forward one-shot learners

Progressive Neural Networks

Matching Networks for One Shot Learning

One-shot Learning with Memory-Augmented Neural Networks

Deep Residual Learning for Image Recognition

A Roadmap Towards Machine Intelligence

Universal Value Function Approximators

Distilling the Knowledge in a Neural Network

Causal inference by using invariant prediction: identification and confidence intervals

Curriculum learning of multiple tasks

Efficient Representations for Lifelong Learning and Autoencoding

Learning and Transferring Mid-level Image Representations Using Convolutional Neural Networks

An Empirical Investigation of Catastrophic Forgeting in Gradient-Based Neural Networks

Learning Factored Representations in a Deep Mixture of Experts

ELLA: An Efficient Lifelong Learning Algorithm

Duality in Quadratic Programming...

Horde: a scalable real-time architecture for learning knowledge from unsupervised sensorimotor interaction

A Survey on Transfer Learning

Toward an Architecture for Never-Ending Language Learning

A theory of learning from different domains

Zero-shot Learning with Semantic Output Codes

Learning to detect unseen object classes by between-class attribute transfer

Curriculum learning

A Bayesian network framework for relational shape matching

Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond

A Model of Inductive Bias Learning

Catastrophic forgetting in connectionist networks

Lifelong Learning Algorithms

Multitask Learning

CHILD: A First Step Towards Continual Learning

Is Learning The n-th Thing Any Easier Than Learning The First?

Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory.

A lifelong learning perspective for mobile robot control

Connectionist models of recognition memory: constraints imposed by learning and forgetting functions.

Learning how to learn

Lifelong Learning with Weighted Majority Votes

Causal and statistical learning

Deep Sequential Neural Networks

Learning Multiple Layers of Features from Tiny Images

The mnist database of handwritten digits

Continual learning in reinforcement environments

Catastrophic Interference in Connectionist Networks: The Sequential Learning Problem

Non-iid input data : the continuum of data is not iid with respect to any ﬁxed probability distribution P ( X, T, Y ) since, once tasks switch, a whole sequence of examples from the new task

Field of Study

Venue Information

Name

Type

URL

Alternate Names