PyTorchKR Papers

최신 논문

총 2,786편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2406.15766

Continual Learning with Diffusion-based Generative Replay for Industrial Streaming Data

The Industrial Internet of Things (IIoT) integrates interconnected sensors and devices to support industrial applications, but its dynamic environments pose challenges related to data drift. Considering the limited resources and the need to effectively adapt models to new data distributions, this paper introduces a Continual Learning (CL) approach, i.e., Distillation-based Self-Guidance (DSG), to address challenges presented by industrial streaming data via a novel generative replay mechanism. DSG utilizes knowledge distillation to transfer knowledge from the previous diffusion-based generator to the updated one, improving both the stability of the generator and the quality of reproduced data, thereby enhancing the mitigation of catastrophic forgetting. Experimental results on CWRU, DSA, and WISDM datasets demonstrate the effectiveness of DSG. DSG outperforms the state-of-the-art baseline in accuracy, demonstrating improvements ranging from 2.9% to 5.0% on key datasets, showcasing its potential for practical industrial applications.

arXiv:2406.16437

Theory on Mixture-of-Experts in Continual Learning

Continual learning (CL) has garnered significant attention because of its ability to adapt to new tasks that arrive over time. Catastrophic forgetting (of old tasks) has been identified as a major issue in CL, as the model adapts to new tasks. The Mixture-of-Experts (MoE) model has recently been shown to effectively mitigate catastrophic forgetting in CL, by employing a gating network to sparsify and distribute diverse tasks among multiple experts. However, there is a lack of theoretical analysis of MoE and its impact on the learning performance in CL. This paper provides the first theoretical results to characterize the impact of MoE in CL via the lens of overparameterized linear regression tasks. We establish the benefit of MoE over a single expert by proving that the MoE model can diversify its experts to specialize in different tasks, while its router learns to select the right expert for each task and balance the loads across all experts. Our study further suggests an intriguing fact that the MoE in CL needs to terminate the update of the gating network after sufficient training rounds to attain system convergence, which is not needed in the existing MoE studies that do not consider the continual task arrival. Furthermore, we provide explicit expressions for the expected forgetting and overall generalization error to characterize the benefit of MoE in the learning performance in CL. Interestingly, adding more experts requires additional rounds before convergence, which may not enhance the learning performance. Finally, we conduct experiments on both synthetic and real datasets to extend these insights from linear models to deep neural networks (DNNs), which also shed light on the practical algorithm design for MoE in CL.

arXiv:2406.14596

VLM Agents Generate Their Own Memories: Distilling Experience into Embodied Programs of Thought

Large-scale LLMs and VLMs excel at few-shot learning but require high-quality examples. We introduce In-Context Abstraction Learning (ICAL), which iteratively refines suboptimal trajectories into high-quality data with optimized actions and detailed reasoning. Given an inefficient demonstration, a VLM corrects actions and annotates causal relationships, object states, subgoals, and task-relevant visuals, forming "programs of thought." With human feedback, these programs are improved as the agent executes them in a similar environment. The resulting examples, used as prompt context or fine-tuning data, significantly boost decision-making while reducing human feedback needs. ICAL surpasses state-of-the-art in TEACh (dialogue-based instruction following), VisualWebArena (multimodal web agents), and Ego4D (egocentric video action anticipation). In TEACh, combining fine-tuning and retrieval on ICAL examples outperforms raw human demonstrations and expert examples, achieving a 17.5% increase in goal-condition success. In VisualWebArena, retrieval-augmented GPT-4V with ICAL improves task success rate 1.6x over GPT-4V, while fine-tuning Qwen2-VL achieves a 2.8x improvement. In Ego4D, ICAL outperforms few-shot GPT-4V and remains competitive with supervised models. Overall, ICAL scales 2x better than raw human demonstrations and reduces manual prompt engineering.

arXiv:2406.15042공식 코드 · Papers with Code

Behaviour Distillation

Dataset distillation aims to condense large datasets into a small number of synthetic examples that can be used as drop-in replacements when training new models. It has applications to interpretability, neural architecture search, privacy, and continual learning. Despite strong successes in supervised domains, such methods have not yet been extended to reinforcement learning, where the lack of a fixed dataset renders most distillation methods unusable. Filling the gap, we formalize behaviour distillation, a setting that aims to discover and then condense the information required for training an expert policy into a synthetic dataset of state-action pairs, without access to expert data. We then introduce Hallucinating Datasets with Evolution Strategies (HaDES), a method for behaviour distillation that can discover datasets of just four state-action pairs which, under supervised learning, train agents to competitive performance levels in continuous control tasks. We show that these datasets generalize out of distribution to training policies with a wide range of architectures and hyperparameters. We also demonstrate application to a downstream task, namely training multi-task agents in a zero-shot fashion. Beyond behaviour distillation, HaDES provides significant improvements in neuroevolution for RL over previous approaches and achieves SoTA results on one standard supervised dataset distillation task. Finally, we show that visualizing the synthetic datasets can provide human-interpretable task insights.

arXiv:2406.13653

Controlling Forgetting with Test-Time Data in Continual Learning

Foundational vision-language models have shown impressive performance on various downstream tasks. Yet, there is still a pressing need to update these models later as new tasks or domains become available. Ongoing Continual Learning (CL) research provides techniques to overcome catastrophic forgetting of previous information when new knowledge is acquired. To date, CL techniques focus only on the supervised training sessions. This results in significant forgetting yielding inferior performance to even the prior model zero shot performance. In this work, we argue that test-time data hold great information that can be leveraged in a self supervised manner to refresh the model's memory of previous learned tasks and hence greatly reduce forgetting at no extra labelling cost. We study how unsupervised data can be employed online to improve models' performance on prior tasks upon encountering representative samples. We propose a simple yet effective student-teacher model with gradient based sparse parameters updates and show significant performance improvements and reduction in forgetting, which could alleviate the role of an offline episodic memory/experience replay buffer.

arXiv:2406.13583

Low-Rank Mixture-of-Experts for Continual Medical Image Segmentation

The primary goal of continual learning (CL) task in medical image segmentation field is to solve the "catastrophic forgetting" problem, where the model totally forgets previously learned features when it is extended to new categories (class-level) or tasks (task-level). Due to the privacy protection, the historical data labels are inaccessible. Prevalent continual learning methods primarily focus on generating pseudo-labels for old datasets to force the model to memorize the learned features. However, the incorrect pseudo-labels may corrupt the learned feature and lead to a new problem that the better the model is trained on the old task, the poorer the model performs on the new tasks. To avoid this problem, we propose a network by introducing the data-specific Mixture of Experts (MoE) structure to handle the new tasks or categories, ensuring that the network parameters of previous tasks are unaffected or only minimally impacted. To further overcome the tremendous memory costs caused by introducing additional structures, we propose a Low-Rank strategy which significantly reduces memory cost. We validate our method on both class-level and task-level continual learning challenges. Extensive experiments on multiple datasets show our model outperforms all other methods.

arXiv:2406.13123공식 코드 · Papers with Code

ViLCo-Bench: VIdeo Language COntinual learning Benchmark

Video language continual learning involves continuously adapting to information from video and text inputs, enhancing a model's ability to handle new tasks while retaining prior knowledge. This field is a relatively under-explored area, and establishing appropriate datasets is crucial for facilitating communication and research in this field. In this study, we present the first dedicated benchmark, ViLCo-Bench, designed to evaluate continual learning models across a range of video-text tasks. The dataset comprises ten-minute-long videos and corresponding language queries collected from publicly available datasets. Additionally, we introduce a novel memory-efficient framework that incorporates self-supervised learning and mimics long-term and short-term memory effects. This framework addresses challenges including memory complexity from long video clips, natural language complexity from open queries, and text-video misalignment. We posit that ViLCo-Bench, with greater complexity compared to existing continual learning benchmarks, would serve as a critical tool for exploring the video-language domain, extending beyond conventional class-incremental tasks, and addressing complex and limited annotation issues. The curated data, evaluations, and our novel method are available at https://github.com/cruiseresearchgroup/ViLCo.

arXiv:2406.12503공식 코드 · Papers with Code

Unsupervised Online Continual Learning for Automatic Speech Recognition

Adapting Automatic Speech Recognition (ASR) models to new domains leads to Catastrophic Forgetting (CF) of previously learned information. This paper addresses CF in the challenging context of Online Continual Learning (OCL), with tasks presented as a continuous data stream with unknown boundaries. We extend OCL for ASR into the unsupervised realm, by leveraging self-training (ST) to facilitate unsupervised adaptation, enabling models to adapt continually without label dependency and without forgetting previous knowledge. Through comparative analysis of various OCL and ST methods across two domain adaptation experiments, we show that UOCL suffers from significantly less forgetting compared to supervised OCL, allowing UOCL methods to approach the performance levels of supervised OCL. Our proposed UOCL extensions further boosts UOCL's efficacy. Our findings represent a significant step towards continually adaptable ASR systems, capable of leveraging unlabeled data across diverse domains.

arXiv:2406.12593

PromptDSI: Prompt-based Rehearsal-free Instance-wise Incremental Learning for Document Retrieval

Differentiable Search Index (DSI) utilizes Pre-trained Language Models (PLMs) for efficient document retrieval without relying on external indexes. However, DSI needs full re-training to handle updates in dynamic corpora, causing significant computational inefficiencies. We introduce PromptDSI, a prompt-based rehearsal-free approach for instance-wise incremental learning document retrieval. PromptDSI attaches prompts to the frozen PLM's encoder of DSI, leveraging its powerful representation to efficiently index new corpora while maintaining a balance between stability and plasticity. We eliminate the initial forward pass of prompt-based continual learning methods that doubles training and inference time. Moreover, we propose a topic-aware prompt pool that employs neural topic embeddings as fixed keys. This strategy ensures diverse and effective prompt usage, addressing the challenge of parameter underutilization caused by the collapse of the query-key matching mechanism. Our empirical evaluations demonstrate that BERT-based PromptDSI matches IncDSI in managing forgetting while improving new corpora performance by more than 4% Hits@10 on NQ320k and upto 3% MRR@10 on MS MARCO 300k.

arXiv:2406.11340

CM2-Net: Continual Cross-Modal Mapping Network for Driver Action Recognition

Driver action recognition has significantly advanced in enhancing driver-vehicle interactions and ensuring driving safety by integrating multiple modalities, such as infrared and depth. Nevertheless, compared to RGB modality only, it is always laborious and costly to collect extensive data for all types of non-RGB modalities in car cabin environments. Therefore, previous works have suggested independently learning each non-RGB modality by fine-tuning a model pre-trained on RGB videos, but these methods are less effective in extracting informative features when faced with newly-incoming modalities due to large domain gaps. In contrast, we propose a Continual Cross-Modal Mapping Network (CM2-Net) to continually learn each newly-incoming modality with instructive prompts from the previously-learned modalities. Specifically, we have developed Accumulative Cross-modal Mapping Prompting (ACMP), to map the discriminative and informative features learned from previous modalities into the feature space of newly-incoming modalities. Then, when faced with newly-incoming modalities, these mapped features are able to provide effective prompts for which features should be extracted and prioritized. These prompts are accumulating throughout the continual learning process, thereby boosting further recognition performances. Extensive experiments conducted on the Drive&Act dataset demonstrate the performance superiority of CM2-Net on both uni- and multi-modal driver action recognition.

arXiv:2406.10785공식 코드 · Papers with Code

ShareLoRA: Parameter Efficient and Robust Large Language Model Fine-tuning via Shared Low-Rank Adaptation

In this paper, we introduce \textbf{Share}d \textbf{Lo}w \textbf{R}ank \textbf{A}daptation (ShareLoRA), a Large Language Model (LLM) fine-tuning technique that balances parameter efficiency, adaptability, and robustness without compromising performance. By strategically sharing the low-rank weight matrices across different layers, ShareLoRA achieves 44\% to 96\% reduction in trainable parameters compared to standard LoRA, alongside a substantial decrease in memory overhead. This efficiency gain scales with model size, making ShareLoRA particularly advantageous for resource-constrained environments. Importantly, ShareLoRA not only maintains model performance but also exhibits robustness in both classification and generation tasks across diverse models, including RoBERTa, GPT-2, and LLaMA series (1, 2, and 3). It consistently outperforms LoRA in zero-shot, few-shot, and continual fine-tuning scenarios, achieving up to 1.2\% average accuracy improvement, and enhanced generalization across domains. In continual learning settings, ShareLoRA achieves 1.2\% higher accuracy on GSM8K, 0.6\% on HumanEval, and 0.5\% on both MMLU and MMLU-Pro. Our results demonstrate that ShareLoRA supports high-quality fine-tuning while offering strong generalization and continual adaptation across various model scales and diverse tasks.

arXiv:2406.11721

Zero-Shot Generalization during Instruction Tuning: Insights from Similarity and Granularity

Understanding alignment techniques begins with comprehending zero-shot generalization brought by instruction tuning, but little of the mechanism has been understood. Existing work has largely been confined to the task level, without considering that tasks are artificially defined and, to LLMs, merely consist of tokens and representations. This line of research has been limited to examining transfer between tasks from a task-pair perspective, with few studies focusing on understanding zero-shot generalization from the perspective of the data itself. To bridge this gap, we first demonstrate through multiple metrics that zero-shot generalization during instruction tuning happens very early. Next, we investigate the facilitation of zero-shot generalization from both data similarity and granularity perspectives, confirming that encountering highly similar and fine-grained training data earlier during instruction tuning, without the constraints of defined "tasks", enables better generalization. Finally, we propose a more grounded training data arrangement method, Test-centric Multi-turn Arrangement, and show its effectiveness in promoting continual learning and further loss reduction. For the first time, we show that zero-shot generalization during instruction tuning is a form of similarity-based generalization between training and test data at the instance level. We hope our analysis will advance the understanding of zero-shot generalization during instruction tuning and contribute to the development of more aligned LLMs. Our code is released at https://github.com/HBX-hbx/dynamics_of_zero-shot_generalization.

arXiv:2406.09935

Forgetting Order of Continual Learning: Examples That are Learned First are Forgotten Last

Catastrophic forgetting poses a significant challenge in continual learning, where models often forget previous tasks when trained on new data. Our empirical analysis reveals a strong correlation between catastrophic forgetting and the learning speed of examples: examples learned early are rarely forgotten, while those learned later are more susceptible to forgetting. We demonstrate that replay-based continual learning methods can leverage this phenomenon by focusing on mid-learned examples for rehearsal. We introduce Goldilocks, a novel replay buffer sampling method that filters out examples learned too quickly or too slowly, keeping those learned at an intermediate speed. Goldilocks improves existing continual learning algorithms, leading to state-of-the-art performance across several image classification tasks.

arXiv:2406.09052공식 코드 · Papers with Code

Data-Free Generative Replay for Class-Incremental Learning on Imbalanced Data

Continual learning is a challenging problem in machine learning, especially for image classification tasks with imbalanced datasets. It becomes even more challenging when it involves learning new classes incrementally. One method for incremental class learning, addressing dataset imbalance, is rehearsal using previously stored data. In rehearsal-based methods, access to previous data is required for either training the classifier or the generator, but it may not be feasible due to storage, legal, or data access constraints. Although there are many rehearsal-free alternatives for class incremental learning, such as parameter or loss regularization, knowledge distillation, and dynamic architectures, they do not consistently achieve good results, especially on imbalanced data. This paper proposes a new approach called Data-Free Generative Replay (DFGR) for class incremental learning, where the generator is trained without access to real data. In addition, DFGR also addresses dataset imbalance in continual learning of an image classifier. Instead of using training data, DFGR trains a generator using mean and variance statistics of batch-norm and feature maps derived from a pre-trained classification model. The results of our experiments demonstrate that DFGR performs significantly better than other data-free methods and reveal the performance impact of specific parameter settings. DFGR achieves up to 88.5% and 46.6% accuracy on MNIST and FashionMNIST datasets, respectively. Our code is available at https://github.com/2younis/DFGR

arXiv:2406.09370

Data-dependent and Oracle Bounds on Forgetting in Continual Learning

In continual learning, knowledge must be preserved and re-used between tasks, maintaining good transfer to future tasks and minimizing forgetting of previously learned ones. While several practical algorithms have been devised for this setting, there have been few theoretical works aiming to quantify and bound the degree of Forgetting in general settings. We provide both data-dependent and oracle upper bounds that apply regardless of model and algorithm choice, as well as bounds for Gibbs posteriors. We derive an algorithm based on our bounds and demonstrate empirically that our approach yields tight bounds on forgetting for several continual learning problems.

arXiv:2406.09384

Reflecting on the State of Rehearsal-free Continual Learning with Pretrained Models

With the advent and recent ubiquity of foundation models, continual learning (CL) has recently shifted from continual training from scratch to the continual adaptation of pretrained models, seeing particular success on rehearsal-free CL benchmarks (RFCL). To achieve this, most proposed methods adapt and restructure parameter-efficient finetuning techniques (PEFT) to suit the continual nature of the problem. Based most often on input-conditional query-mechanisms or regularizations on top of prompt- or adapter-based PEFT, these PEFT-style RFCL (P-RFCL) approaches report peak performances; often convincingly outperforming existing CL techniques. However, on the other end, critical studies have recently highlighted competitive results by training on just the first task or via simple non-parametric baselines. Consequently, questions arise about the relationship between methodological choices in P-RFCL and their reported high benchmark scores. In this work, we tackle these questions to better understand the true drivers behind strong P-RFCL performances, their placement w.r.t. recent first-task adaptation studies, and their relation to preceding CL standards such as EWC or SI. In particular, we show: (1) P-RFCL techniques relying on input-conditional query mechanisms work not because, but rather despite them by collapsing towards standard PEFT shortcut solutions. (2) Indeed, we show how most often, P-RFCL techniques can be matched by a simple and lightweight PEFT baseline. (3) Using this baseline, we identify the implicit bound on tunable parameters when deriving RFCL approaches from PEFT methods as a potential denominator behind P-RFCL efficacy. Finally, we (4) better disentangle continual versus first-task adaptation, and (5) motivate standard RFCL techniques s.a. EWC or SI in light of recent P-RFCL methods.

arXiv:2406.07590공식 코드 · Papers with Code

StreamFP: Learnable Fingerprint-guided Data Selection for Efficient Stream Learning

Stream Learning (SL) requires models that can quickly adapt to continuously evolving data, posing significant challenges in both computational efficiency and learning accuracy. Effective data selection is critical in SL to ensure a balance between information retention and training efficiency. Traditional rule-based data selection methods struggle to accommodate the dynamic nature of streaming data, highlighting the necessity for innovative solutions that effectively address these challenges. Recent approaches to handling changing data distributions face challenges that limit their effectiveness in fast-paced environments. In response, we propose StreamFP, a novel approach that uniquely employs dynamic, learnable parameters called fingerprints to enhance data selection efficiency and adaptability in stream learning. StreamFP optimizes coreset selection through its unique fingerprint-guided mechanism for efficient training while ensuring robust buffer updates that adaptively respond to data dynamics, setting it apart from existing methods in stream learning. Experimental results demonstrate that StreamFP outperforms state-of-the-art methods by achieving accuracy improvements of 15.99%, 29.65%, and 51.24% compared to baseline models across varying data arrival rates, alongside a training throughput increase of 4.6x.

arXiv:2406.08431

Diffusion Soup: Model Merging for Text-to-Image Diffusion Models

We present Diffusion Soup, a compartmentalization method for Text-to-Image Generation that averages the weights of diffusion models trained on sharded data. By construction, our approach enables training-free continual learning and unlearning with no additional memory or inference costs, since models corresponding to data shards can be added or removed by re-averaging. We show that Diffusion Soup samples from a point in weight space that approximates the geometric mean of the distributions of constituent datasets, which offers anti-memorization guarantees and enables zero-shot style mixing. Empirically, Diffusion Soup outperforms a paragon model trained on the union of all data shards and achieves a 30% improvement in Image Reward (.34 $\to$ .44) on domain sharded data, and a 59% improvement in IR (.37 $\to$ .59) on aesthetic data. In both cases, souping also prevails in TIFA score (respectively, 85.5 $\to$ 86.5 and 85.6 $\to$ 86.8). We demonstrate robust unlearning -- removing any individual domain shard only lowers performance by 1% in IR (.45 $\to$ .44) -- and validate our theoretical insights on anti-memorization using real data. Finally, we showcase Diffusion Soup's ability to blend the distinct styles of models finetuned on different shards, resulting in the zero-shot generation of hybrid styles.

arXiv:2406.08171

Continuous fake media detection: adapting deepfake detectors to new generative techniques

Generative techniques continue to evolve at an impressively high rate, driven by the hype about these technologies. This rapid advancement severely limits the application of deepfake detectors, which, despite numerous efforts by the scientific community, struggle to achieve sufficiently robust performance against the ever-changing content. To address these limitations, in this paper, we propose an analysis of two continuous learning techniques on a Short and a Long sequence of fake media. Both sequences include a complex and heterogeneous range of deepfakes generated from GANs, computer graphics techniques, and unknown sources. Our study shows that continual learning could be important in mitigating the need for generalizability. In fact, we show that, although with some limitations, continual learning methods help to maintain good performance across the entire training sequence. For these techniques to work in a sufficiently robust way, however, it is necessary that the tasks in the sequence share similarities. In fact, according to our experiments, the order and similarity of the tasks can affect the performance of the models over time. To address this problem, we show that it is possible to group tasks based on their similarity. This small measure allows for a significant improvement even in longer sequences. This result suggests that continual techniques can be combined with the most promising detection methods, allowing them to catch up with the latest generative techniques. In addition to this, we propose an overview of how this learning approach can be integrated into a deepfake detection pipeline for continuous integration and continuous deployment (CI/CD). This allows you to keep track of different funds, such as social networks, new generative tools, or third-party datasets, and through the integration of continuous learning, allows constant maintenance of the detectors.

arXiv:2406.06811

Learning Continually by Spectral Regularization

Loss of plasticity is a phenomenon where neural networks can become more difficult to train over the course of learning. Continual learning algorithms seek to mitigate this effect by sustaining good performance while maintaining network trainability. We develop a new technique for improving continual learning inspired by the observation that the singular values of the neural network parameters at initialization are an important factor for trainability during early phases of learning. From this perspective, we derive a new spectral regularizer for continual learning that better sustains these beneficial initialization properties throughout training. In particular, the regularizer keeps the maximum singular value of each layer close to one. Spectral regularization directly ensures that gradient diversity is maintained throughout training, which promotes continual trainability, while minimally interfering with performance in a single task. We present an experimental analysis that shows how the proposed spectral regularizer can sustain trainability and performance across a range of model architectures in continual supervised and reinforcement learning settings. Spectral regularization is less sensitive to hyperparameters while demonstrating better training in individual tasks, sustaining trainability as new tasks arrive, and achieving better generalization performance.