Towards LifeSpan Cognitive Systems
Authors: Yu Wang, Chi Han, Tongtong Wu, Xiaoxin He, Wangchunshu Zhou, Nafis Sadeq, Xiusi Chen, Zexue He, Wei Wang, Gholamreza Haffari, Heng Ji, Julian McAuley
TMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Theoretical | In this paper, we propose the concept of a Life Span Cognitive System (LSCS), which aims to manage rapid, incremental learning while retaining the ability to recall previous experiences. To achieve LSCS, we argue there are two major challenges: (1) Abstraction and Experiences Merging, (2) Long-Term Retention and Accurate Recalling. Existing technologies have the potential to partially solve these challenges. Based on our defined metric, Storage Complexity, for saving past experiences (E), we categorize existing technologies into four distinct classes. We summarize various technologies within each category and discuss the strengths and weaknesses of the methods in each category. We argue that achieving an LSCS requires a nuanced approach integrating multiple strategies to address the above two challenges. To this end, we propose a conceptual instantiation for LSCS. We hope to provide insights into LSCS and encourage further efforts. |
| Researcher Affiliation | Collaboration | Yu Wang1 , Chi Han2 , Tongtong Wu3 , Xiaoxin He4 , Wangchunshu Zhou5, Nafis Sadeq1, Xiusi Chen2,6, Zexue He1,7, Wei Wang6, Gholamreza Haffari3, Heng Ji2, Julian Mc Auley1 1UCSD, 2UIUC, 3Monash, 4NUS, 5AIWaves, 6UCLA, 7MIT-IBM Correspondence to EMAIL. |
| Pseudocode | No | The paper describes the proposed instantiation's processes ('Absorbing Experiences' and 'Generating Responses') in textual form and through a diagram (Figure 2), but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement or a direct link to the source code for the methodology described in this work. It mentions open-source tools or implementations by other researchers (e.g., 'Open-source tools like Lang Chain and Llama Index', 'open-source implementations (Munkhdalai et al., 2024; Yang et al., 2024)') but not for its own proposed conceptual framework. |
| Open Datasets | No | The paper is theoretical and conceptual, proposing a framework. It does not conduct experiments using specific datasets. It mentions benchmarks (e.g., 'Bench (Zhang et al., 2024d) or Loong Bench (Wang et al., 2024b)') in the context of challenges, but these are not used for evaluation within this paper. |
| Dataset Splits | No | The paper is theoretical and does not present experimental results or use datasets in a way that would require specifying splits for reproduction. |
| Hardware Specification | No | The paper is theoretical and does not describe any experimental setup, therefore no specific hardware specifications are provided. |
| Software Dependencies | No | The paper is theoretical and does not present any experimental results. Therefore, it does not list specific software dependencies with version numbers. |
| Experiment Setup | No | The paper is theoretical and conceptual, proposing a framework for a Life Span Cognitive System. It does not include any experimental results, hyperparameters, or training configurations. |