Rehearsal-Free Continual Learning over Small Non-I.I.D. Batches (2019-07-08T00:00:00.000000Z)

TL;DR

A novel continual learning protocol based on the CORe50 benchmark is introduced and two rehearsal-free continual learning techniques are proposed, CWR* and AR1*, that can learn effectively even in the challenging case of nearly 400 small non-i.i.d. incremental batches.

Abstract

Robotic vision is a field where continual learning can play a significant role. An embodied agent operating in a complex environment subject to frequent and unpredictable changes is required to learn and adapt continuously. In the context of object recognition, for example, a robot should be able to learn (without forgetting) objects of never before seen classes as well as improving its recognition capabilities as new instances of already known classes are discovered. Ideally, continual learning should be triggered by the availability of short videos of single objects and performed on-line on on-board hardware with fine-grained updates. In this paper, we introduce a novel continual learning protocol based on the CORe50 benchmark and propose two rehearsal-free continual learning techniques, CWR* and AR1*, that can learn effectively even in the challenging case of nearly 400 small non-i.i.d. incremental batches. In particular, our experiments show that AR1* can outperform other state-of-the-art rehearsal-free techniques by more than 15% accuracy in some cases, with a very light and constant computational and memory overhead across training batches.

Authors

Vincenzo Lomonaco

2 papers

D. Maltoni

2 papers

Lorenzo Pellegrini

1 papers

TL;DR

Abstract

Authors

References38 items

Lifelong Machine Learning with Deep Streaming Linear Discriminant Analysis

Fine-Grained Continual Learning

Continual Learning for Robotics

EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks

Continual Learning with Deep Architectures

Memory Efficient Experience Replay for Streaming Learning

Continuous Learning in Single-Incremental-Task Scenarios

Lifelong Learning of Spatiotemporal Representations With Dual-Memory Recurrent Self-Organization

The limits and potentials of deep learning for robotics

Continual Lifelong Learning with Neural Networks: A Review

MobileNetV2: Inverted Residuals and Linear Bottlenecks

Are we Done with Object Recognition? The iCub robot's Perspective

Measuring Catastrophic Forgetting in Neural Networks

Gradient Episodic Memory for Continual Learning

Lifelong machine learning: a paradigm for continuous learning

CORe50: a New Dataset and Benchmark for Continuous Object Recognition

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

Continual Learning Through Synaptic Intelligence

Batch Renormalization: Towards Reducing Minibatch Dependence in Batch-Normalized Models

Overcoming catastrophic forgetting in neural networks

iCaRL: Incremental Classifier and Representation Learning

Fine-Tuning Deep Neural Networks in Continuous Learning Scenarios

Xception: Deep Learning with Depthwise Separable Convolutions

Learning without Forgetting

Semi-supervised tuning from temporal coherence

Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

Adam: A Method for Stochastic Optimization

Caffe: Convolutional Architecture for Fast Feature Embedding

The stability-plasticity dilemma: investigating the continuum from catastrophic forgetting to age-limited learning effects

Incremental Learning by Message Passing in Hierarchical Temporal Memory

Catastrophic forgetting in connectionist networks

Neural networks

Catastrophic Forgetting, Rehearsal and Pseudorehearsal

The Utilization of Multiple Measurements in Problems of Biological Classification

Batch ReNormalization Implementation, 2017

Teaching iCub to recognize objects using deep Convolutional Neural Networks

propose a selective weight freeze for the CNN models adopting Depth-Wise Separable Convolutions (Section 2.3)

On the Fly Object Recognition on the R1, Your Personal Hu-manoid - IIT

Field of Study

Journal Information

Name

Page