Marginal Replay vs Conditional Replay for Continual Learning

By Timothée Lesort, Alexander Rainer Tassilo Gepperth, Andrei Stoian, David Filliat. 2018

38 citations recorded by the source corpus.

We present a new replay-based method of continual classification learning that we term "conditional replay" which generates samples and labels together by sampling from a distribution conditioned on the class. We compare conditional replay to another replay-based continual learning paradigm (which we term "marginal replay") that generates samples independently of their class and assigns labels in a separate step. The main improvement in conditional replay is that labels for generated samples need not be inferred, which reduces the margin for error in complex continual classification learning tasks. We demonstrate the effectiveness of this approach using novel and standard benchmarks constructed from MNIST and FashionMNIST data, and compare to the regularization-based EWC method.

Subfields: Artificial Intelligence, Civil and Structural Engineering, Computer Vision and Pattern Recognition, Meta & Continual Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Civil and Structural Engineering

In Computer Vision and Pattern Recognition

In Meta & Continual Learning

Citations in the corpus

DisCoRL: Continual Reinforcement Learning via Policy Distillation (2019, methodology): “Another method we also exploit is to ﬁrst learn a state representation [Lesort et al., 2018] to compress the observation into a low dimensional embedding and secondly learn the policy on top of this representation.”
Continual Learning on Noisy Data Streams via Self-Purified Replay (2021, neutral): “This problem is often coined as continual learning, for which three representative approaches have been proposed [57, 67, 20] including replay [52, 29, 66, 73, 70, 44], regularization [38, 91, 3], and expansion techniques [71, 89].”
Meta-Consolidation for Continual Learning (2020, background): “One line of work [88, 38, 47, 2, 16, 71, 56, 13, 79, 42] constrains the parameters of the deep network trained on Task A to not change much while learning a new Task B, while another - replay-based methods - store [63, 48, 14, 66, 9, 3, 4, 22] or generate [74, 81, 46, 45] examples of previous tasks ”
Imbalanced Continual Learning with Partitioning Reservoir Sampling (2020, methodology): “We adopt the replay-based approach [56,35,3,76,73,49] to tackle continual learning, which explicitly stores the past experiences into a memory or a generative model, and rehearses them back with the new input samples.”
Online Continual Learning with Maximally Interfered Retrieval (2019, background): “fective alternative to the constraint optimization approach [8, 5]. Rehearsal methods usually play random samples from a buffer, or pseudo samples from a generative model trained on the previous data [37, 23, 24]. These works showed promising results in the ofﬂine incremental tasks setting and recen”
Continual Learning for Robotics (2019, methodology): “, by first learning a state representation (Lesort et al., 2018) and later performing policy learning (Finn et al.”
Defining Benchmarks for Continual Few-Shot Learning (2020, methodology): “The problem of continual learning (CL), also called lifelong learning, has been considered since the beginnings of artificial intelligence and it remains an open challenge in robotics (Lesort et al., 2019c) and machine learning (Parisi et al., 2019).”
Continual Learning in Deep Networks: an Analysis of the Last Layer (2021, background): “Continual learning research field studied many strategies, such as rehearsal [26, 9, 2, 3, 52, 19, 6], generative replay [46, 53, 28, 30, 40], dynamic architecture [16, 42, 32] or regularization [22, 55, 41, 44].”
Understanding Continual Learning Settings with Data Distribution Drift Analysis (2021, background): “…evaluation in continual learning often referred to as “ﬁnal performance” (Chaudhry et al., 2019; Kirkpatrick et al., 2017; Shin et al., 2017; Lesort et al., 2019c), however in a more general scope, it evaluates the current performance of an algorithm on a test task at a particular point…”
Online Learned Continual Compression with Adaptive Quantization Modules (2019, background): “Although CL has been studied in generative modeling (Ramapuram et al., 2017; Lesort et al., 2018; Zhai et al., 2019; Lesort et al., 2019) and reinforcement learning (Kirk-patrick et al., 2017; Fernando et al., 2017; Riemer et al., 2018), supervised learning is still the standard for evalua-Online…”
Multi-Domain Multi-Task Rehearsal for Lifelong Learning (2020, background): “…at the layers below the rehearsal layer and leaves the layers above free to learn at full pace; (3) by building generative model to synthesize data (Pseudo-rehearsal) (Shen et al. 2020; van de Ven and Tolias 2018; Lesort et al. 2019), the knowledge can be saved as parameters rather than data.”
Online Learned Continual Compression with Stacked Quantization Module (2019, background): “Although CL has been studied in generative modeling (Ramapuram et al., 2017; Lesort et al., 2018; Zhai et al., 2019; Lesort et al., 2019) and reinforcement learning (Kirkpatrick et al., 2017; Fernando et al., 2017; Riemer et al., 2018), supervised learning is still the standard for evaluation of…”
Sequoia: A Software Framework to Unify Continual Learning Research (2021, methodology): “In CSL, this assumption differentiates class-incremental learning from domain-incremental learning (van de Ven & Tolias, 2019).”