PyTorchKR Papers

최신 논문

총 573편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:1612.00796

Overcoming catastrophic forgetting in neural networks

The ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Neural networks are not, in general, capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of connectionist models. We show that it is possible to overcome this limitation and train networks that can maintain expertise on tasks which they have not experienced for a long time. Our approach remembers old tasks by selectively slowing down learning on the weights important for those tasks. We demonstrate our approach is scalable and effective by solving a set of classification tasks based on the MNIST hand written digit dataset and by learning several Atari 2600 games sequentially.

arXiv:1611.07725공식 코드 · Papers with Code

iCaRL: Incremental Classifier and Representation Learning

A major open problem on the road to artificial intelligence is the development of incrementally learning systems that learn about more and more concepts over time from a stream of data. In this work, we introduce a new training strategy, iCaRL, that allows learning in such a class-incremental way: only the training data for a small number of classes has to be present at the same time and new classes can be added progressively. iCaRL learns strong classifiers and a data representation simultaneously. This distinguishes it from earlier works that were fundamentally limited to fixed data representations and therefore incompatible with deep learning architectures. We show by experiments on CIFAR-100 and ImageNet ILSVRC 2012 data that iCaRL can learn many classes incrementally over a long period of time where other strategies quickly fail.

arXiv:1705.00744공식 코드 · Papers with Code

A Strategy for an Uncompromising Incremental Learner

Multi-class supervised learning systems require the knowledge of the entire range of labels they predict. Often when learnt incrementally, they suffer from catastrophic forgetting. To avoid this, generous leeways have to be made to the philosophy of incremental learning that either forces a part of the machine to not learn, or to retrain the machine again with a selection of the historic data. While these hacks work to various degrees, they do not adhere to the spirit of incremental learning. In this article, we redefine incremental learning with stringent conditions that do not allow for any undesirable relaxations and assumptions. We design a strategy involving generative models and the distillation of dark knowledge as a means of hallucinating data along with appropriate targets from past distributions. We call this technique, phantom sampling.We show that phantom sampling helps avoid catastrophic forgetting during incremental learning. Using an implementation based on deep neural networks, we demonstrate that phantom sampling dramatically avoids catastrophic forgetting. We apply these strategies to competitive multi-class incremental learning of deep neural networks. Using various benchmark datasets and through our strategy, we demonstrate that strict incremental learning could be achieved. We further put our strategy to test on challenging cases, including cross-domain increments and incrementing on a novel label space. We also propose a trivial extension to unbounded-continual learning and identify potential for future development.

arXiv:1705.08690

Continual Learning with Deep Generative Replay

Attempts to train a comprehensive artificial intelligence capable of solving multiple tasks have been impeded by a chronic problem called catastrophic forgetting. Although simply replaying all previous data alleviates the problem, it requires large memory and even worse, often infeasible in real world applications where the access to past data is limited. Inspired by the generative nature of hippocampus as a short-term memory system in primate brain, we propose the Deep Generative Replay, a novel framework with a cooperative dual model architecture consisting of a deep generative model ("generator") and a task solving model ("solver"). With only these two models, training data for previous tasks can easily be sampled and interleaved with those for a new task. We test our methods in several sequential learning settings involving image classification tasks.

arXiv:1803.00159

A Class-Incremental Learning Method Based on One Class Support Vector Machine

A method based on one class support vector machine (OCSVM) is proposed for class incremental learning. Several OCSVM models divide the input space into several parts. Then, the 1VS1 classifiers are constructed for the confuse part by using the support vectors. During the class incremental learning process, the OCSVM of the new class is trained at first. Then the support vectors of the old classes and the support vectors of the new class are reused to train 1VS1 classifiers for the confuse part. In order to bring more information to the certain support vectors, the support vectors are at the boundary of the distribution of samples as much as possible when the OCSVM is built. Compared with the traditional methods, the proposed method retains the original model and thus reduces memory consumption and training time cost. Various experiments on different datasets also verify the efficiency of the proposed method.

arXiv:1806.02942공식 코드 · Papers with Code

SupportNet: solving catastrophic forgetting in class incremental learning with support data

A plain well-trained deep learning model often does not have the ability to learn new knowledge without forgetting the previously learned knowledge, which is known as catastrophic forgetting. Here we propose a novel method, SupportNet, to efficiently and effectively solve the catastrophic forgetting problem in the class incremental learning scenario. SupportNet combines the strength of deep learning and support vector machine (SVM), where SVM is used to identify the support data from the old data, which are fed to the deep learning model together with the new data for further training so that the model can review the essential information of the old data when learning the new information. Two powerful consolidation regularizers are applied to stabilize the learned representation and ensure the robustness of the learned model. We validate our method with comprehensive experiments on various tasks, which show that SupportNet drastically outperforms the state-of-the-art incremental learning methods and even reaches similar performance as the deep learning model trained from scratch on both old and new data. Our program is accessible at: https://github.com/lykaust15/SupportNet

arXiv:2411.16073v4cs.LG공식 코드 · Papers with Code

연속 학습을 위한 Soft-TransFormers

Soft-TransFormers for Continual Learning

잘 초기화된 복권 티켓 가설(Well-initialized Lottery Ticket Hypothesis, WLTH)에서 영감을 받아, 우리는 Soft-TransFormers(Soft-TF)를 제안한다. 이는 고정된 사전학습 트랜스포머를 선택된 셀프 어텐션(self-attention) 레이어의 query, key, value, output 투영(projection)에 대한 실수값 곱셈 마스크로 구성된 과제별 소프트 서브네트워크를 통해 적응시키는 연속 학습(continual learning) 프레임워크이다. 마스크는 1로 초기화되므로 최적화는 정확히 사전학습 해법에서 시작하며, 마스크 공간 경사하강법은 본질적으로 백본의 지배적 경로를 조절하는 방향으로 편향된다. 우리는 표준 볼록-리ipschitz 가정 하에서, 수렴 속도와 마스크만 사용하는 미세조정의 매개변수 편향(drift) 모두가 사전학습 가중치에서 과제 최적 구성까지의 거리에 의해 제어됨을 증명한다. 이러한 제한된 편향은 두 가지 성질을 만든다. 백본과 과제별 마스크가 덮어쓰이지 않으므로 망각은 구조적으로 제거된다. 또한 모든 과제 서브네트워크가 공유 사전학습 해법 근처에 머물기 때문에, 잘못된 마스크라도 거의 범용적인 함수에 가까운 함수를 평가하므로 과제 추론 오류는 대체로 무해하며, 클래스 증분(class-incremental) 정확도는 과제 추론 신뢰성과 분리된다. 플러그인으로 Soft-TF는 L2P, DualPrompt, HiDe-Prompt, NoRGa와 결합되며, 과제 키 매칭, 엔트로피-그래디언트 기준, 또는 학습된 과제 식별 분류기로 마스크를 선택한다. Split-CIFAR100, Split-ImageNet-R, CUB-200, 5-Datasets 등 클래스 증분 학습 벤치마크에서 Soft-TF는 유사한 학습 가능 매개변수 예산에서 프롬프트 기반, 어댑터 기반, LoRA 스타일 기준선들을 일관되게 능가하면서, 수정되지 않은 백본과 동일한 추론 비용을 유지한다.

arXiv:2507.08922v3stat.ML

연속 학습에 대한 베이지안 접근법: 개요

The Bayesian Approach to Continual Learning: An Overview

연속 학습(continual learning)은 학습자가 순차적 시간 단계에 걸쳐 마주치는 다양한 태스크로부터 지식을 지속적으로 축적하는 온라인 패러다임이다. 중요하게도, 학습자는 과거에 축적한 학습 경험을 망각하지 않으면서, 또한 처음부터 재학습할 필요 없이 지식을 확장하고 갱신하도록 요구된다. 순차적 특성과 인간의 사고 방식과의 유사성 덕분에, 연속 학습은 딥러닝 모델의 실세계 문제 적용 범위를 확장하는 데 현재 걸림돌이 되는 여러 도전 과제를 해결할 기회를 제공한다. 순차적으로 도착하는 데이터로 학습자를 지속적으로 갱신해야 한다는 점은 연속 학습과 베이지안 추론(Bayesian inference) 사이의 본질적인 일치성을 만들어 내는데, 베이지안 추론은 이전 데이터에서 획득한 지식을 완전히 망각하지 않으면서 새로운 데이터가 주어졌을 때 모델의 사전 신념(prior belief)을 지속적으로 갱신하기 위한 원리적 플랫폼을 제공한다. 본 서베이는 태스크 증분 학습(task-incremental learning)과 클래스 증분 학습(class-incremental learning)이라는 베이지안 연속 학습의 다양한 설정을 고찰한다. 먼저 연속 학습의 정의와 베이지안 설정, 그리고 도메인 적응(domain adaptation), 전이학습(transfer learning), 메타학습(meta-learning)과 같은 관련 분야와의 연관성을 논의하는 것으로 시작한다. 이어서 베이지안 연속 학습 패러다임에 속하는 알고리즘들을 포괄적으로 분류하는 택소노미를 제시한다. 동시에 현재의 최신 기법을 분석하고, 현재까지 가장 주목할 만한 베이지안 연속 학습 알고리즘 일부를 상세히 살펴본다. 아울러 연속 학습과 발달심리학 간의 연관성을 조명하고, 두 분야 사이의 유사점을 소개한다. 마지막으로 현재의 도전 과제를 논의하고, 베이지안 연속 학습의 미래 연구 방향으로 잠재적 영역을 제시하며 마무리한다.

arXiv:2607.01630v1cs.CV

DRDN: 처음부터 학습하는 ViT 클래스 증분 학습을 위한 분리 표현 동적 네트워크

DRDN: Decoupled Representation Dynamic Network for From-Scratch ViT Class-Incremental Learning

클래스 증분 학습(CIL)에서 분류 지도만으로는 공유 백본 표현이 충분히 보존되지 않는 문제와 태스크 간 혼동을 해결하기 위해, DRDN은 백본에만 그래디언트를 라우팅하는 마스크 이미지 모델링(MIM)과 계층적 태스크 토큰 확장을 결합한 분리 표현 동적 네트워크를 제안한다. 외부 사전학습 없이 처음부터 학습하는 ViT CIL 설정에서 CIFAR100-B0 10단계 기준 77.19% 평균 정확도로 DyTox 대비 3.53포인트 향상을 달성하며, MIM 디코더는 추론 시 추가 연산 없이 학습 중에만 활성화된다.

arXiv:2606.29465v1cs.LG

클래스 증분 학습을 위한 프로토타입 잠재 세계 모델 리플레이

Prototype Latent World Model Replay for Class-Incremental Learning

원본 샘플 없이 이전 클래스를 보존하는 클래스 증분 학습(class-incremental learning) 문제를 해결하기 위해, 동결된 ImageNet 사전학습 인코더의 안정적인 잠재 상태 공간에 각 클래스를 프로토타입 중심 분포로 저장하고 이를 리플레이하는 Prototype Latent World Model Replay 프레임워크를 제안한다. Split CIFAR-100 벤치마크에서 원본 예시 저장 없이 Inc5/Inc10/Inc20 설정의 LastAcc를 각각 31.64%, 37.06%, 43.10%로 끌어올려 단순 파인튜닝 대비 큰 폭의 성능 향상을 달성했다.

arXiv:2606.27095v1cs.LG

효율적인 장기 콜드 스타트 지속 학습을 위한 데이터 없는 리저voir 특징

Data-Free Reservoir Features for Efficient Long-Horizon Cold-Start Continual Learning

리플레이·외부 사전학습·큰 초기 태스크 없이 클래스가 점점 늘어나는 콜드 스타트 exemplar-free 클래스 증분 학습(CS-EFCIL)을 위해, CIRCLE은 이미지 데이터에 적합(fit)되지 않은 고정 BiRC2D 스타일 2차원 리저voir(reservoir) 특징과 스트리밍 선형 판별 분석(SLDA) 헤드 앙상블을 결합해 리플레이·태스크 경계 정보·백본 역전파 없이 샘플 단위로 학습한다. CIFAR-100, TinyImageNet, ImageNet-Subset, ImageNet-1k에서 10–20 태스크 분할에서는 경쟁력을 보이고 50–500 분할에서는 강한 CS-EFCIL baseline을 크게 능가하며, 백본을 반복 학습하는 drift 보상 방법보다 훨씬 빠르다.

arXiv:2511.11421v2cs.CV

BOFA: Bridge-Layer Orthogonal Low-Rank Fusion for CLIP-Based Class-Incremental Learning

Class-Incremental Learning (CIL) aims to continually learn new categories without forgetting previously acquired knowledge. Vision-language models such as CLIP offer strong transferable representations via multi-modal supervision, making them promising for CIL. However, applying CLIP to CIL poses two major challenges: (1) adapting to downstream tasks often requires additional learnable modules, increasing model complexity and susceptibility to forgetting; and (2) while multi-modal representations offer complementary strengths, existing methods have yet to fully realize their potential in effectively integrating visual and textual modalities. To address these issues, we propose BOFA (Bridge-layer Orthogonal Fusion for Adaptation), a novel framework for CIL. BOFA confines all model adaptation exclusively to CLIP's existing cross-modal bridge-layer, thereby adding no extra parameters or inference cost. To prevent forgetting within this layer, it leverages Orthogonal Low-Rank Fusion, a mechanism that constrains parameter updates to a low-rank ``safe subspace" mathematically constructed to be orthogonal to past task features. This ensures stable knowledge accumulation without data replay. Furthermore, BOFA employs a cross-modal hybrid prototype that synergizes stable textual prototypes with visual counterparts derived from our stably adapted bridge-layer, enhancing classification performance. Extensive experiments on standard benchmarks show that BOFA achieves superior accuracy and efficiency compared to existing methods.

arXiv:2407.06136v4cs.CV공식 코드 · Papers with Code

Mamba-FSCIL: Dynamic Adaptation with Selective State Space Model for Few-Shot Class-Incremental Learning

Few-shot class-incremental learning (FSCIL) aims to incrementally learn novel classes from limited examples while preserving knowledge of previously learned classes. Existing methods face a critical dilemma: static architectures rely on a constant parameter space to learn from data that arrive sequentially, making them prone to overfitting to the current session, while dynamic architectures continually expand the parameter space, leading to increased complexity. In this study, we explore the potential of Selective State Space Models (SSMs) for FSCIL. Mamba leverages its input-dependent parameters to dynamically adjust its processing patterns and generate content-aware scan patterns without session-wise projector expansion. This enables it to configure distinct processing for base and novel classes, helping preserve existing knowledge while adapting to new ones. To leverage Mamba's potential for FSCIL, we design two key modules: First, we propose a dual selective SSM projector that generates input-conditioned state-space parameters from intermediate features for dynamic adaptation. The dual design structurally decouples base and novel-class processing, employing a frozen base branch to maintain stable base-class features and a dynamic incremental branch that adaptively learns distinctive feature shifts for novel classes. Second, we develop a class-sensitive selective scan mechanism to guide dynamic adaptation of the incremental branch. It reduces the disruption to base-class representations caused by training on novel data, and meanwhile, encourages the selective scan to perform in distinct patterns between base and novel classes. Extensive experiments on miniImageNet, CIFAR-100, and CUB-200 demonstrate that Mamba-FSCIL achieves state-of-the-art performance.