EMBER: Autonomous Cognitive Behaviour from Learned Spiking Neural Network Dynamics in a Hybrid LLM Architecture
TLDR
EMBER is a hybrid LLM-SNN architecture that enables autonomous cognitive behavior by using learned SNN dynamics to trigger and shape LLM actions.
Key contributions
- A hybrid architecture (EMBER) integrates a 220,000-neuron SNN with an LLM, placing the LLM as a replaceable reasoning engine.
- SNN uses STDP, a 4-layer hierarchy, E/I balance, and reward-modulated learning for persistent associative memory.
- A novel z-score standardized top-k population code encodes text embeddings into the SNN, retaining 82.2% discrimination.
- The SNN autonomously triggers and shapes LLM actions via lateral propagation during idle periods, without external prompts.
Why it matters
This paper introduces a novel approach to AI cognition by allowing a biologically-inspired SNN to autonomously control an LLM, moving beyond simple retrieval augmentation. It demonstrates a system that can initiate complex actions based on internal learned dynamics, leading to more adaptive and less scripted AI behavior.
Original Abstract
We present (Experience-Modulated Biologically-inspired Emergent Reasoning), a hybrid cognitive architecture that reorganises the relationship between large language models (LLMs) and memory: rather than augmenting an LLM with retrieval tools, we place the LLM as a replaceable reasoning engine within a persistent, biologically-grounded associative substrate. The architecture centres on a 220,000-neuron spiking neural network (SNN) with spike-timing-dependent plasticity (STDP), four-layer hierarchical organisation (sensory/concept/category/meta-pattern), inhibitory E/I balance, and reward-modulated learning. Text embeddings are encoded into the SNN via a novel z-score standardised top-k population code that is dimension-independent by construction, achieving 82.2\% discrimination retention across embedding dimensionalities. We show that STDP lateral propagation during idle operation can trigger and shape LLM actions without external prompting or scripted triggers: the SNN determines when to act and what associations to surface, while the LLM selects the action type and generates content. In one instance, the system autonomously initiated contact with a user after learned person-topic associations fired laterally during an 8-hour idle period. From a clean start with zero learned weights, the first SNN-triggered action occurred after only 7 conversational exchanges (14 messages).
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.