Social learning highlights that learning agents improve not in isolation, but through interaction and structured knowledge exchange with others. When introduced into machine learning, this principle gives rise to social machine learning (SML), where multiple agents collaboratively learn by sharing abstracted knowledge. Federated learning (FL) provides a natural collaboration substrate for this paradigm, yet existing heterogeneous FL approaches often rely on sharing model parameters or intermediate representations, which may expose sensitive information and incur additional overhead. In this work, we propose SoHip (Social Hippocampus Memory Learning), a memory-centric social machine learning framework that enables collaboration among heterogeneous agents via memory sharing rather than model sharing. SoHip abstracts each agent's individual short-term memory from local representations, consolidates it into individual long-term memory through a hippocampus-inspired mechanism, and fuses it with collectively aggregated long-term memory to enhance local prediction. Throughout the process, raw data and local models remain on-device, while only lightweight memory are exchanged. We provide theoretical analysis on convergence and privacy preservation properties. Experiments on two benchmark datasets with seven baselines demonstrate that SoHip consistently outperforms existing methods, achieving up to 8.78% accuracy improvements.
Figures & tables
Figure 1: Memory-centric social machine learning framework.
Figure 2: Overview of SoHip . SoHip operates sequentially by (1) abstracting individual short-term memory from local representations, (2) consolidating it into individual long-term memory via a hippocampus-inspired mechanism, (3) fusing it with collective long-term memory for enhanced prediction, and (4) aggregating updated individual long-term memories to form an updated collective memory.
Method
CIFAR-100
ImageNet
N=100
N=200
N=300
N=100
N=200
N=300
Standalone
53.59 ± 0.48
47.35 ± 0.52
42.92 ± 0.61
35.36 ± 0.57
28.29 ± 0.63
25.94 ± 0.69
FedProto ( Tan et al., 2022b )
53.54 ± 0.44
45.25 ± 0.58
43.90 ± 0.55
34.43 ± 0.60
28.05 ± 0.66
24.66 ± 0.72
FedSSA ( Yi et al., 2024c )
47.39 ± 0.63
42.98 ± 0.69
41.04 ± 0.74
29.99 ± 0.71
25.90 ± 0.77
22.51 ± 0.81
FedRAL ( Yi et al., 2025a )
53.32 ± 0.46
45.56 ± 0.51
44.62 ± 0.59
35.31 ± 0.55
27.82 ± 0.61
25.82 ± 0.67
FedKD ( Wu et al., 2022 )
35.39 ± 0.82
29.86 ± 0.91
26.56 ± 0.96
24.54 ± 0.88
17.08 ± 0.94
13.37 ± 1.02
Table 1: Average test accuracy (%) under the pathological label-skew partition with client participation rate C=10% and varying number of clients N . Results are reported as mean ± standard deviation over three runs. Best results in each column are in bold.
Figure 3: Test accuracy curves under pathological label-skew settings ( C=10% ) on CIFAR-100 (top) and Tiny-ImageNet (bottom) with varying numbers of agents, where SoHip consistently achieves faster convergence and higher final accuracy across all settings.
Figure 4: Impact of non-IID degree under pathological label-skew by varying classes per agent and practical label-skew by Dirichlet partition with concentration α ( ⋆ is default in Table 1 ).
Figure 5: Impact of memory dimension m on SoHip .
Figure 6: Ablation results of SoHip .
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Notation
Description
N
Total number of agents (clients).
i
Index of an agent, i∈{1,…,N} .
t
Communication round index.
Di
Local private dataset of agent i .
Fi
Heterogeneous feature extractor of agent i .
Hi
Local classifier (prediction head) of agent i .
Appendix
Table 2: Summary of notations used in SoHip .
Figure 7: Sensitivity analysis of SoHip with respect to agent participation rate and learning rate.
Motivated by recursive self-improvement (RSI), we study how LLM agents can learn through external memory without updating model parameters. Extracting useful experience from a single attempt is difficult: flawed reasoning can produce success, while poor execution can obscure a sound strategy. Preserving its value over time is also difficult: our accumulation baseline without consolidation initially improves but eventually performs below a no-memory baseline. Our key idea is to compare attempts at the same task to identify reusable decisions and refine when they apply. We introduce Hippo, a memory framework inspired by selective hippocampal encoding and replay that learns from these comparisons when shortcomings are diagnosed. It compares the resulting experience with existing memory to merge overlaps and resolve conflicts through explicit applicability conditions. This allows prior guidance to be revised without storing every lesson as a separate entry. Experiments on SWE-bench Verified, WebArena, and Mind2Web show performance gains in all three settings and substantially fewer retained experiences than the accumulation baseline. The gains extend across the evaluated base models of varying capabilities, indicating that memory-based adaptation remains useful even for highly capable models.
Yutong Song, Jiang Wu, Yuxuan Ren +2
University of California, Irvine · Independent Researcher · Rice University +1
Designing effective memory mechanisms is crucial for advancing LLM-driven Multi-Agent Systems (MAS), helping agents learn together and perform better over time. While recent work has led to strong cooperation skills, most methods still use flat, unstructured memories, which easily get filled with noise and erase differences between agents. To address this, we introduce the concept of collective-individual memory synergy and propose CoMem, an architecture that unifies both private experience and shared knowledge for multi-agent learning. CoMem features:(i) Private Experience Sedimentation, which lets each agent keep and update its own useful memories over time;(ii) Collective Wisdom Curation, which carefully selects only widely proven ideas to be shared among agents;(iii)Parallel Dual-Stream Retrieval, which allows agents to draw both from their own memory and the group's wisdom, using clustering to ensure diversity.Experiments on ALFWorld and PDDL benchmarks show that CoMem achieves strong overall performance and robustly avoids memory pollution.
Chengxin Yu, Zhaoxin Fan, Faguo Wu +3
Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing, School of Artificial Intelligence, Beihang University · Beijing academy of blockchain and edge computing · National University of Defense Technology +1
LLM-based agents face challenges in personalized interactive tasks due to heterogeneous, multi-typed, and implicitly constrained long-term traces. Existing memory mechanisms struggle with accurate routing and retrieval, especially on-device where personalization is critical. Most methods use single-vector representations, blurring type distinctions and relational structure. We propose HGP, a hybrid graph memory framework. HGP employs a lightweight self-enhancement classifier for personalized memory routing and constructs episodic, semantic, and procedural memories as graphs. It also extracts working memory as a state trajectory to capture current state and implicit constraints, ensuring reliable decision-making. The classifier reduces large-model calls, enabling on-device deployment, while graph storage enables accurate retrieval and incremental user profile refinement. Experiments on two benchmarks show that on PAL-Set solution selection, HGP achieves an S-score of 35.58, nearly 7 points above the strongest baseline. Code and data are at https://github.com/Ouan6/HGP-.git.
Ran Zhou, Xueming Han, Jiaheng Liu +4
School of Intelligence Science and Technology, Nanjing University · Jiutian Research