The Synapse Context Engine (SCE) synthesizes ideas from multiple research traditions. This document provides complete citations and explains the conceptual connections.
Important Note: Citations are divided into two categories:
- Core Implementation — Concepts directly implemented in SCE's architecture
- Conceptual Inspiration — Ideas that influenced design thinking but are not directly implemented
These concepts are directly implemented in SCE's current architecture.
Citation:
Hebb, D. O. (1949). The Organization of Behavior: A Neuropsychological Theory. Wiley.
ISBN: 978-0805843002 (reprint edition)
Note: This is a classic text widely available in university libraries and used bookstores. Digital copies may be found through academic library systems.
What SCE Implements:
Hebbian weight learning: "neurons that fire together wire together." When nodes co-activate during context construction, their synaptic weights strengthen through the learning mechanism described in the concept paper (Section 4.1, Equation 3).
Citation:
Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82(6), 407-428.
DOI:
https://doi.org/10.1037/h0077081
What SCE Implements:
The core activation propagation mechanism (Equation 1 in concept paper). Query nodes inject energy that spreads through the hypergraph with decay factors (γ), thresholds (θ), and depth limits—directly implementing spreading activation dynamics.
Citation:
Anderson, J. R., Bothell, D., Byrne, M. D., Douglass, S., Lebiere, C., & Qin, Y. (2004). An integrated theory of the mind. Psychological Review, 111(4), 1036-1060.
DOI:
https://doi.org/10.1037/0033-295X.111.4.1036
What SCE Implements:
Activation thresholds for memory retrieval. Only nodes exceeding activation threshold θ (e.g., 0.3) participate in context synthesis, preventing noise and ensuring relevance—directly adapted from ACT-R's activation-based memory access.
Citation:
Carbonell, J., & Goldstein, J. (1998). The use of MMR, diversity-based reranking for reordering documents and producing summaries. Proceedings of SIGIR '98, 335-336.
URL:
https://www.cs.cmu.edu/~jgc/publication/The_Use_MMR_Diversity_Based_LTMIR_1998.pdf
What SCE Implements:
Maximal Marginal Relevance pruning (implemented in pruneWithMMR function). Calculates node selection score as λ * Relevance - (1-λ) * Redundancy, balancing relevance to query with diversity from already-selected context—the standard information-theoretic approximation for context selection.
Citation:
Cover, T. M., & Thomas, J. A. (2006). Elements of Information Theory (2nd ed.). Wiley-Interscience.
DOI:
https://doi.org/10.1002/047174882X
What SCE Implements:
Information-theoretic context compression principles (Section 5.2, Equation 6-8). While practical implementation uses embedding-based approximations, the theoretical foundation is KL divergence and information gain for selecting maximally informative, minimally redundant nodes.
Citation:
Chung, F. R. K. (1997). Spectral Graph Theory. American Mathematical Society.
URL:
https://bookstore.ams.org/cbms-92/
What SCE Implements:
Heat diffusion on graphs (Equation 4 & 5 in concept paper). Implemented via Graph Laplacian Diffusion (applyHeatDiffusion), where "heat" (activation energy) flows from active nodes to connected neighbors over time, simulating temporal decay and recency bias handling.
These works influenced SCE's design philosophy but are not directly implemented in the current architecture. They represent aspirational directions or metaphorical framing.
Citation:
O'Keefe, J., & Nadel, L. (1978). The Hippocampus as a Cognitive Map. Oxford University Press.
ISBN: 978-0198572060
Inspirational Contribution:
The metaphor of memory as a "navigable cognitive map" rather than address-based storage influenced SCE's framing. However, SCE does not implement hippocampal models—this is conceptual inspiration, not architectural mimicry.
Citation:
McClelland, J. L., McNaughton, B. L., & O'Reilly, R. C. (1995). Why there are complementary learning systems in the hippocampus and neocortex: Insights from the successes and failures of connectionist models of learning and memory. Psychological Review, 102(3), 419-457.
DOI:
https://doi.org/10.1037/0033-295X.102.3.419
Inspirational Contribution:
The distinction between fast episodic encoding and slow consolidation influenced thinking about rapid graph updates vs. gradual weight adjustment. However, SCE does not implement a formal dual-system architecture—this is metaphorical inspiration.
Citation:
Laird, J. E., Newell, A., & Rosenbloom, P. S. (1987). SOAR: An architecture for general intelligence. Artificial Intelligence, 33(1), 1-64.
DOI:
https://doi.org/10.1016/0004-3702(87)90050-6
Inspirational Contribution:
SOAR's symbolic memory navigation and problem-space exploration informed SCE's bounded recursive traversal approach. The conceptual similarity is in "navigating knowledge spaces" rather than direct implementation.
Citation:
Berge, C. (1973). Graphs and Hypergraphs. North-Holland Mathematical Library, Volume 6.
ISBN: 978-0720424492
What SCE Implements:
True hypergraph support with multi-way edges. SCE maintains both pairwise synapses (source→target) and hyperedges (connecting multiple nodes simultaneously). When any member of a hyperedge activates, energy distributes to all connected nodes (clique activation)—implementing the core hypergraph concept of atomic multi-way relationships.
Citation:
Estrada, E., & Hatano, N. (2008). Communicability in complex networks. Physical Review E, 77(3), 036111.
DOI:
https://doi.org/10.1103/PhysRevE.77.036111
Inspirational Contribution:
Informed thinking about how information flows through graph structures. SCE does not calculate formal communicability metrics, but the concept influenced design considerations for activation propagation.
Core Implementation citations represent actual working code in SCE. Conceptual Inspiration citations acknowledge intellectual debt without claiming direct implementation.
SCE is a practical systems architecture built by an AI engineer, not an academic research project. These citations serve two purposes:
- Give credit where ideas originated
- Help researchers understand the conceptual lineage
The goal is working software informed by these ideas, adapted for the realities of LLMs, hypergraphs, and real-time interaction—not faithful reproduction of academic models.
This is a living document. Additional papers and theoretical connections will be added as the research progresses and new relevant work is identified.
These papers represent alternative approaches to AI memory and context management that SCE builds upon or contrasts with.
Citation:
Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., ... & Kiela, D. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. Advances in Neural Information Processing Systems, 33, 9459-9474.
arXiv:
https://arxiv.org/abs/2005.11401
Relevance to SCE:
RAG is the dominant paradigm for incorporating external knowledge into LLMs via vector similarity retrieval. SCE was designed to address RAG's limitations in persistent systems: contextual fragmentation, flat relevance scoring, and lack of relational structure.
Citation:
Park, J. S., O'Brien, J. C., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023). Generative agents: Interactive simulacra of human behavior. arXiv preprint arXiv:2304.03442.
arXiv:
https://arxiv.org/abs/2304.03442
Relevance to SCE:
Introduced episodic memory and reflection mechanisms for long-lived AI agents. While innovative, Generative Agents still rely on flat memory representations. SCE explores whether graph-based memory can provide richer contextual coherence for similar use cases.
Citation:
Packer, C., Fang, V., Patil, S. G., Wooders, K., & Stoica, I. (2023). MemGPT: Towards LLMs as operating systems. arXiv preprint arXiv:2310.08560.
arXiv:
https://arxiv.org/abs/2310.08560
Relevance to SCE:
MemGPT treats memory as hierarchical storage with explicit paging and context management. Both MemGPT and SCE address the problem of bounded context windows, but take different architectural approaches: MemGPT uses OS-style memory management, SCE uses graph-based associative retrieval.
- SCE Concept Paper: docs/blueprints/sce_initial_concept.pdf
- Architecture Notes: docs/notes/architecture_notes.md
- GitHub Repository: https://github.com/sasus-dev/synapse-context-engine
This work is licensed under a Creative Commons Attribution 4.0 International License.
