A Neural Dirichlet Process Mixture Model for Task-Free Continual Learning

By Soochan Lee, Junsoo Ha, Dongsu Zhang, Gunhee Kim. 2020

242 citations recorded by the source corpus.

Despite the growing interest in continual learning, most of its contemporary works have been studied in a rather restricted setting where tasks are clearly distinguishable and task boundaries are known during training. However, if our goal is to develop an algorithm that learns as humans do, this setting is far from realistic and it is essential to develop a methodology that works in a task-free manner. Meanwhile, among several branches of continual learning, expansion-based methods have the advantage of eliminating catastrophic forgetting by allocating new resource to learn new data. In this work, we propose an expansion-based approach for task-free continual learning for the first time. Our model, named Continual Neural Dirichlet Process Mixture (CN-DPM), consists of a set of neural network experts that are in charge of a subset of the data. CN-DPM expands the number of experts in a principled way under the Bayesian nonparametric framework. With extensive experiments, we show that our model successfully performs task-free continual learning for both discriminative and generative tasks such as image classification and image generation.

Subfields: Probabilistic & Bayesian Methods, 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 Probabilistic & Bayesian Methods

In Meta & Continual Learning

Citations in the corpus

A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, methodology): “CN-DPM [80] adopts Dirichlet process mixture models to construct a growing number of neural experts, while a concurrent work [269] derives such mixture models from a probabilistic meta-learner.”
The Ideal Continual Learner: An Agent That Never Forgets (2023, background): “…and more challenging case of task-agnostic or task-free continual learning, where in the training phase the learner has no access to task identities (Zeno et al., 2018; 2021; Aljundi et al., 2019a; Lee et al., 2020; Jin et al., 2021; Wang et al., 2022c; Pourcel et al., 2022; Ye & Bors, 2022).”
CODA-Prompt: COntinual Decomposed Attention-Based Prompting for Rehearsal-Free Continual Learning (2022, background): “One group of approaches expands a model’s architecture as new tasks are encountered; these are highly effective for applications where a model growing with tasks is practical [16, 35, 39, 43, 54].”
Task-Free Continual Learning via Online Discrepancy Distance Learning (2022, neutral): “This issue can be addressed by using Dirichlet processes by adding new components while freezing all previously learnt units, in a model called the Continual Neural Dirichlet Process Mixture (CN-DPM), [24].”
Improving Task-free Continual Learning by Distributionally Robust Memory Evolution (2022, methodology): “This learning scenario has become increasingly important due to the broader application scenarios, and more challenging problem nature (Aljundi et al., 2019b; Lee et al., 2020).”
Class-Incremental Learning by Knowledge Distillation with Adaptive Feature Consolidation (2022, background): “Continual neural Dirichlet process mixture [24] estimates the joint likelihood of a new sample and its label by which it decides whether an additional network component is required to handle the new sample.”
Continual Learning via Local Module Composition (2021, methodology): “Continual learning methods typically address the problem of forgetting through parameter regularization [46, 70, 99], replay [89, 78, 3, 73, 54, 12, 97, 39, 81, 13] or dynamic architectures (and MoEs) [83, 87, 52, 85, 53, 41].”
Online Continual Learning on Class Incremental Blurry Task Configuration with Anytime Inference (2021, methodology): “For a more realistic and practical CL setup, considering real-world scenarios, we strictly adhere to the online and task-free CL setup (Lee et al., 2019; Losing et al., 2018).”
Continual Learning on Noisy Data Streams via Self-Purified Replay (2021, methodology): “There have been three main branches to train a model from continual data streams: regularization [51, 19, 38, 3], expansion [71, 89, 43], and replay [52, 9, 10, 69, 34].”
Always Be Dreaming: A New Approach for Data-Free Class-Incremental Learning (2021, methodology): “For example, methods which expand a model’s architecture as new tasks are encountered are highly effective for applications where a model growing with tasks is practical [14, 34, 38, 42, 50, 53].”
Same State, Different Task: Continual Reinforcement Learning without Interference (2021, background): “By contrast, expansion approaches add new neural resources to enable learning new tasks while preserving components for specific tasks (Rusu et al. 2016; Lee et al. 2020).”
Supervised Contrastive Replay: Revisiting the Nearest Class Mean Classifier in Online Class-Incremental Continual Learning (2021, background): “Parameter-isolation methods bypass interference by allocating different parameters to each task [38, 32, 60].”
Posterior Meta-Replay for Continual Learning (2021, methodology): “Other methods like CN-DPM (Lee et al., 2020) use Bayes’ rule for task inference on the joint p(X, C), where C is a discrete condition such as the task identity.”
Online Continual Learning in Image Classification: An Empirical Survey (2021, methodology): “CN-DPM has terrible performance in this setting because it is very sensitive to hyperparameters, and the method cannot find the hyperparameter set that works in this setting within the same tuning budget as other methods.”
Meta-Consolidation for Continual Learning (2020, methodology): “We compare with a recent bayesian continual learing method, CN-DPM [44], for completeness of our work.”
Imbalanced Continual Learning with Partitioning Reservoir Sampling (2020, background): “• No test-time task labels: Our approach does not require explicit task labels during both training and test phase, often coined as task-free continual learning in [3,47].”
Continual Prototype Evolution: Learning Online from Non-Stationary Data Streams (2020, methodology): “CIFAR setups use a slim version of Resnet18 [19] with a 1k memory size for CIFAR10 [4, 29], and 5k for CIFAR100 [31].”
Online Class-Incremental Continual Learning with Adversarial Shapley Value (2020, methodology): “Online Class-Incremental Learning Following the recent CL literature (Aljundi et al. 2019a,b; Lee et al. 2020; Chrysakis and Moens 2020), we consider the online supervised class-incremental learning setting where a model needs to learn new classes continually from an online data stream (each sample…”