Optimal Foraging in Memory Retrieval: Evaluating Random Walks and Metropolis-Hastings Sampling in Modern Semantic Spaces

TL;DR

This work reframes semantic memory retrieval as foraging in a high-dimensional embedding space and tests whether simple retrieval processes can reproduce Marginal Value Theorem (MVT)–driven switching patterns. Using OpenAI's text-embedding-large-3 to create a 550-item animal space augmented with descriptive sentences, it compares a temperature-weighted random walk and a Metropolis-Hastings sampler, evaluating performance with inter-item retrieval-time (IRT) metrics. The key finding is that random walks on richly structured embeddings can approximate human-foraging dynamics, while MH sampling does not consistently outperform the simpler approach; descriptive embeddings are crucial for aligning with human data. The results challenge the view that more complex, adaptive sampling is always necessary, supporting Hills (2012) over Abbott (2015) and suggesting that appropriate representations can yield near-optimal foraging with straightforward processes.

Abstract

Human memory retrieval often resembles ecological foraging where animals search for food in a patchy environment. Optimal foraging means following the Marginal Value Theorem (MVT), in which individuals exploit a patch of semantically related concepts until it becomes less rewarding and then switch to a new cluster. While human behavioral data suggests foraging-like patterns in semantic fluency tasks, it remains unclear whether modern high-dimensional embedding spaces provide representations that allow algorithms to match observed human behavior. Using state-of-the-art embeddings and prior semantic fluency data, I find that random walks on these embedding spaces produce results consistent with optimal foraging and the MVT. Surprisingly, introducing Metropolis-Hastings sampling, an adaptive algorithm expected to model strategic acceptance and rejection of new clusters, does not produce results consistent with human behavior. These findings challenge the assumption that more complex sampling mechanisms inherently lead to better cognitive models of memory retrieval. Instead, they show that appropriately structured embeddings, even with simple sampling, can produce near-optimal foraging dynamics. This supports the perspective of Hills (2012) rather than Abbott (2015), demonstrating that modern embeddings can approximate human memory foraging without relying on complex acceptance criteria.

Figures (4)

• Figure 1: t-SNE, cosine similarity heatmaps and additive clustering model visualizations. Observe that the descriptive embeddings produce better structured clusters with respect to the latent clusters.
• Figure 2: Empirical results from Abbott et al. (2015).
• Figure 3: Results for non-descriptive embeddings.
• Figure 4: Results for descriptive embeddings.