ExplicitLM: Decoupling Knowledge from Parameters via Explicit Memory Banks

TL;DR

ExplicitLM tackles knowledge staleness and interpretability in large language models by introducing a million-entry external Memory Bank ($N=10^6$, $L=16$) that stores human-readable facts for inspection and editing. It employs a two-stage differentiable retrieval with product-key decomposition, reducing search complexity from $ ext{O}(N\cdot|I|)$ to $ ext{O}(\\sqrt{N}\cdot|I|)$, and partitions knowledge into frozen explicit facts and learnable implicit patterns with Exponential Moving Average updates for stability. The approach achieves up to $43.67\%$ improvement on knowledge-intensive tasks and up to $3.62\times$ gains in low-data regimes (10k), with retrieval success strongly predicting accuracy; even perfect retrieval yields additional gains. By enabling transparent, updatable knowledge management, ExplicitLM addresses reliability and interpretability concerns that challenge traditional RAG systems while maintaining competitive performance.

Abstract

Large language models suffer from knowledge staleness and lack of interpretability due to implicit knowledge storage across entangled network parameters, preventing targeted updates and reasoning transparency. We propose ExplicitLM, a novel architecture featuring a million-scale external memory bank storing human-readable knowledge as token sequences, enabling direct inspection and modification. We design a differentiable two-stage retrieval mechanism with efficient coarse-grained filtering via product key decomposition (reducing complexity from $\mathcal{O}(N \cdot |I|)$ to $\mathcal{O}(\sqrt{N} \cdot |I|)$) and fine-grained Gumbel-Softmax matching for end-to-end training. Inspired by dual-system cognitive theory, we partition knowledge into frozen explicit facts (20%) and learnable implicit patterns (80%), maintained through Exponential Moving Average updates for stability. ExplicitLM achieves up to 43.67% improvement on knowledge-intensive tasks versus standard Transformers, with 3.62$\times$ gains in low-data regimes (10k samples). Analysis shows strong correlations between memory retrieval and performance, with correct predictions achieving 49% higher hit rates. Unlike RAG systems with frozen retrieval, our jointly optimized architecture demonstrates that interpretable, updatable models can maintain competitive performance while providing unprecedented knowledge transparency.

Figures (4)

• Figure 1: Overall architecture of ExplicitLM. The blue region shows the multi-layer transformer blocks. The gray region represents the shared Memory Bank accessed by all layers, where each layer can retrieve knowledge via the Memory Retrieval Mechanism (Section \ref{['Memory Retrieval Mechanism']}) from Explicit Knowledge (green) or Implicit Knowledge (yellow) partitions. The orange region shows a sample knowledge entry from the Memory Bank—a sequence of token indices of length $L$ directly convertible to human-readable text.
• Figure 2: ExplicitLM architecture with memory retrieval mechanism. In Stage 1, both query and key vectors are partitioned along the embedding dimension into two components for efficient retrieval. In Stage 2, cosine similarity is computed between the query and candidate knowledge entries, with the highest-scoring entry selected for retrieval.
• Figure 3: Layer-wise memory hit rates for Relation Reasoning across varying training data volumes. Semi-transparent regions indicate hit rates for correctly predicted samples, while opaque regions show hit rates for incorrect predictions. Red annotations display the hit rate differential between correct and incorrect predictions at each layer.
• Figure 4: Performance comparison across different freeze rates. The bar chart shows accuracy values under various experimental conditions with different training set sizes. The line plot indicates the relative performance improvement (in percentage) of our method compared to the baseline at different freeze rates for each training set size.