Abstract:When an agent acquires new information, ideally it would immediately be capable of using that information to understand its environment. This is not possible using conventional deep neural networks, which suffer from catastrophic forgetting when they are incrementally updated, with new knowledge overwriting established representations. A variety of approaches have been developed that attempt to mitigate catastrophic forgetting in the incremental batch learning scenario, where a model learns from a series of large collections of labeled samples. However, in this setting, inference is only possible after a batch has been accumulated, which prohibits many applications. An alternative paradigm is online learning in a single pass through the training dataset on a resource constrained budget, which is known as streaming learning. Streaming learning has been much less studied in the deep learning community. In streaming learning, an agent learns instances one-by-one and can be tested at any time, rather than only after learning a large batch. Here, we revisit streaming linear discriminant analysis, which has been widely used in the data mining research community. By combining streaming linear discriminant analysis with deep learning, we are able to outperform both incremental batch learning and streaming learning algorithms on both ImageNet ILSVRC-2012 and CORe50, a dataset that involves learning to classify from temporally ordered samples.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of catastrophic forgetting encountered by deep neural networks (DNN) in the process of lifelong learning. Specifically, the paper focuses on how to effectively train deep neural networks in the streaming learning scenario, enabling the model to learn new data in real - time under resource - constrained conditions and not forget the previously learned knowledge. #### Main problem description: 1. **Catastrophic Forgetting**: Traditional deep neural networks will experience catastrophic forgetting during incremental updates, that is, when learning new knowledge, the existing knowledge representation will be overwritten, resulting in a sharp decline in performance on old tasks. 2. **Online Learning under Resource Constraints**: In many practical applications, especially in embedded devices and real - time systems, the model must be able to perform single - pass learning in a data stream, and cannot repeatedly access the data set during this process. In addition, the model also needs to operate under limited computing resources and memory conditions. 3. **Insufficiency of Existing Solutions**: - **Incremental Batch Learning**: Although there are some methods that can alleviate catastrophic forgetting in incremental batch learning, these methods usually require multiple traversals of data batches, which is not feasible in streaming learning. - **Existing Streaming Learning Methods**: Most of the existing streaming learning methods are suitable for low - dimensional data streams, and the training speed is slow, not suitable for embedded applications. #### Solutions proposed in the paper: The paper proposes a method that combines deep learning and Deep Streaming Linear Discriminant Analysis (Deep SLDA) to achieve an efficient, lightweight classifier in the streaming learning scenario. Specifically: - **Deep SLDA**: By applying Streaming Linear Discriminant Analysis (SLDA) to the last layer of a deep convolutional neural network (CNN), the paper shows how to achieve efficient streaming learning by only updating the output - layer parameters without storing previous data. - **Performance Advantages**: Experimental results show that Deep SLDA not only outperforms existing streaming learning algorithms on large - scale image classification data sets such as ImageNet ILSVRC - 2012 and CORe50, but is also faster and more memory - efficient than incremental batch learning methods. #### Summary: The core problem of this paper is to solve the problem of catastrophic forgetting in deep neural networks in the streaming learning scenario. Deep SLDA is proposed as a lightweight and efficient solution that can achieve real - time learning and inference in resource - constrained environments.

Lifelong Machine Learning with Deep Streaming Linear Discriminant Analysis

Deep Streaming Label Learning

STREAMLINE: Streaming Active Learning for Realistic Multi-Distributional Settings

Memory-efficient training with streaming dimensionality reduction

Lifelong Bayesian Learning Machines for Streaming Industrial Big Data

Edge-Cloud Collaborative Streaming Video Analytics with Multi-agent Deep Reinforcement Learning

VERSE: Virtual-Gradient Aware Streaming Lifelong Learning with Anytime Inference

XLDA: Linear Discriminant Analysis for Scaling Continual Learning to Extreme Classification at the Edge

Explainable Lifelong Stream Learning Based on "Glocal" Pairwise Fusion

Implicit Context-aware Learning and Discovery for Streaming Data Analytics

Streaming Deep Reinforcement Learning Finally Works

DeepStreamCE: A Streaming Approach to Concept Evolution Detection in Deep Neural Networks

Streaming Active Deep Forest for Evolving Data Stream Classification

A Unified Replay-based Continuous Learning Framework for Spatio-Temporal Prediction on Streaming Data

Online Boosting Adaptive Learning under Concept Drift for Multistream Classification

Adaptive Deep Models for Incremental Learning

Online Active Learning for Drifting Data Streams

Adaptive Online Incremental Learning for Evolving Data Streams

Online Learning From Incomplete and Imbalanced Data Streams

Streaming Active Learning with Deep Neural Networks

A Deep Learning Framework for Lifelong Machine Learning