Progressive Prompts: Continual Learning for Language Models

Anastasia, Razdaibiedina,; Mao, Yuning; Hou, Rui; Khabsa, Madian; Lewis, Michael A.; Almahairi, Amjad

doi:10.48550/arxiv.2301.12314

Cited by 4 publications

(8 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In Table 2, we compare with the previous state-of-the-art methods: MBPA++ [5], LAMOL [31], IDBR [10] and ProgressivePrompt [25] on full-setting datasets. The results show that C&F achieves the best performances on 4 different orders.…”

Section: Resultsmentioning

confidence: 99%

“…IDBR [10] disentangles text representation space into a task generic space and a task specific space. ProgressivePrompt [25] learns a new soft prompt for each task and sequentially concatenates it with the previously learned ones, while keeping the base model frozen.…”

Section: Methods For Comparisonmentioning

confidence: 99%

“…Architecture-based method focuses on allocating a subset of the model's parameters for each task, optimizing the model exclusively on those parameters. Some allocate and utilize existing parameters to maintain a fixed model size [7,18] and others freeze all model parameters and introduce additional parameters for each new task [25,27]. However, the former imposes a limit on the model's capacity and the latter requires expanding the storage occupied by the model as new tasks appear; 3.…”

Section: Continual Learningmentioning

confidence: 99%

See 2 more Smart Citations

Create and Find Flatness: Building Flat Training Spaces in Advance for Continual Learning

Shi,

Chen,

Zhao

et al. 2023

Frontiers in Artificial Intelligence and Applications

View full text Add to dashboard Cite

Catastrophic forgetting remains a critical challenge in the field of continual learning, where neural networks struggle to retain prior knowledge while assimilating new information. Most existing studies emphasize mitigating this issue only when encountering new tasks, overlooking the significance of the pre-task phase. Therefore, we shift the attention to the current task learning stage, presenting a novel framework, C&F (Create and Find Flatness), which builds a flat training space for each task in advance. Specifically, during the learning of the current task, our framework adaptively creates a flat region around the minimum in the the loss landscape. Subsequently, it finds the parameters’ importance to the current task based on their flatness degrees. When adapting the model to a new task, constraints are applied according to the flatness and a flat space is simultaneously prepared for the impending task. We theoretically demonstrate the consistency between the created and found flatness. In this manner, our framework not only accommodates ample parameter space for learning new tasks but also preserves the preceding knowledge of earlier tasks. Experimental results exhibit C&F’s state-of-the-art performance as a standalone continual learning approach and its efficacy as a framework incorporating other methods. Our work is available at https://github.com/Eric8932/Create-and-Find-Flatness.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Methods For Comparisonmentioning

confidence: 99%

Section: Continual Learningmentioning

confidence: 99%

See 1 more Smart Citation

Create and Find Flatness: Building Flat Training Spaces in Advance for Continual Learning

Shi,

Chen,

Zhao

et al. 2023

Frontiers in Artificial Intelligence and Applications

View full text Add to dashboard Cite

show abstract

“…The model starts with simple prompts and progresses to more challenging ones as it becomes more proficient or as the task requires (Zheng et al 2023) . Progressive prompting allows the model to incrementally build upon its knowledge and skills, leading to more sophisticated output (Razdaibiedina et al 2023) .…”

Section: Progressive Promptingmentioning

confidence: 99%

A Review of Generative Pretrained Multi-step Prompting Schemes –and a New Multi-step Prompting Framework

Cohen,

Aperstein

2024

Preprint

View full text Add to dashboard Cite

This paper reviews the pioneering work related to the generation of a sequence of prompts in Pretrained Language Models (PLMs) to automatically complete a complex writing task. The paper identifies the main categories of automated multi-step prompting and discusses their advantages and disadvantages. The paper also proposes an additional approach for automated multi-step prompting with iterative four main phases: in the first phase a sequence of prompts is geared to elicit the structure of the requested task. This is done by identifying the task’s main parts, the general contents of the different parts and their precedence relations. The structure of each main part is iteratively explored to find secondary and tertiary parts if they exist. A second phase is followed by a sequence of prompts that fill in the primary, secondary, and tertiary parts, and seek to fine-tune the result. The third phase reviews the generated text and criticizes it. Based on this criticism a revision is prepared. The fourth phase is fine-tuning the document, with emphasis on section length and text tone. Future research could deal with the system identification of points where the system should initiate interaction with the user for desired feedback and guidance.

show abstract

“…In rehearsal-based methods, data from previously learned tasks are stored in a rehearsal buffer and used in the current task in addition to the current training set [3,4,33]. A more recent promptbased rehearsal-free approach combines powerful pretrained backbones with learnable prompts that retain the knowledge acquired from the different tasks without modifying the weights of the main backbone, thus avoiding forgetting [52,53,41,37]. Our method gets inspiration from the solutions proposed in the latter methods.…”

Section: Incremental Learningmentioning

confidence: 99%

Multimodal Parameter-Efficient Few-Shot Class Incremental Learning

D’Alessandro¹,

Alonso²,

Calabrés³

et al. 2023

Preprint

View full text Add to dashboard Cite

Few-Shot Class Incremental Learning (FSCIL) is a challenging continual learning task, where limited training examples are available during several learning sessions. To succeed in this task, it is necessary to avoid over-fitting new classes caused by biased distributions in the few-shot training sets. The general approach to address this issue involves enhancing the representational capability of a predefined backbone architecture by adding special modules for backward compatibility with older classes. However, this approach has not yet solved the dilemma of ensuring high classification accuracy over time while reducing the gap between the performance obtained on larger training sets and the smaller ones. In this work, we propose an alternative approach called Continual Parameter-Efficient CLIP (CPE-CLIP) to reduce the loss of information between different learning sessions. Instead of adapting additional modules to address information loss, we leverage the vast knowledge acquired by CLIP in large-scale pre-training and its effectiveness in generalizing to new concepts. Our approach is multimodal and parameter-efficient, relying on learnable prompts for both the language and vision encoders to enable transfer learning across sessions. We also introduce prompt regularization to improve performance and prevent forgetting. Our experimental results demonstrate that CPE-CLIP significantly improves FSCIL performance compared to state-of-the-art proposals while also drastically reducing the number of learnable parameters and training costs.Recent research has focused on solving these problems through various approaches, such as meta-learning [57, 34], regularization techniques [30], or knowledge distillation [38,6,62]. These methods have shown promising results in achieving incremental learning over time with a limited amount of data available. The general approaches consist in enhancing the basic representational capability of a predefined backbone architecture by adding special modules to entail backward compatibility with older classes during learning sessions. These solutions are computationally expensive since they need a large number of iterations in each session to adapt the additional modules to new classes while maintaining backward compatibility. Despite the high computational cost, they still fail to efficiently reduce the gap between the performance obtained on larger training sets

show abstract

Progressive Prompts: Continual Learning for Language Models

Cited by 4 publications

References 16 publications

Create and Find Flatness: Building Flat Training Spaces in Advance for Continual Learning

Create and Find Flatness: Building Flat Training Spaces in Advance for Continual Learning

A Review of Generative Pretrained Multi-step Prompting Schemes –and a New Multi-step Prompting Framework

Multimodal Parameter-Efficient Few-Shot Class Incremental Learning

Contact Info

Product

Resources

About