Bio-inspired optimization ยท Open Source ยท ML ๐Ÿ„

Mycorrhizal network priors
for LLM optimization

Myco-Opt uses the structural properties of fungal networks โ€” evolved solutions to resource-flow optimization under uncertainty โ€” as biological priors that guide transformer hyperparameter search, reducing the evaluation budget.

The problem

Training a large LLM costs millions. Finding the right hyperparameters (depth, experts, learning rate, routing) is still largely trial-and-error โ€” and every evaluation is extremely expensive.

$M

Cost per training run

A single frontier-model run can cost millions of dollars. Hyperparameter search multiplies that cost.

~50%

Budget we aim to save

Our hypothesis: biological priors reach โ‰ฅ Bayesian Optimization quality with โ‰ค50% of the evaluation budget.

600M+

Real observations

GlobalFungi (Nature Sci Data 2020) + MycoDB: open fungal interaction data to derive the priors.

The bio-inspired idea

Mycorrhizal networks connect plants through soil and transport resources. They are networks optimized by millions of years of evolution under uncertainty โ€” the same problem we face when designing AI architectures.

๐Ÿ„ Biological transport networks

We do not claim "fungal intelligence". We use verifiable quantitative topology: modularity, degree distribution, small-world connectivity, diameter and clustering โ€” measured on real open data.

๐Ÿง  Priors, not absolute truths

The mappings constrain the search space toward biologically plausible regions. The evolutionary loop with an LLM optimizer (OPRO-style) corrects the rest.

Fungal metrics โ†’ Hyperparameters

Each structural property of the network maps naturally onto the transformer design space.

Network metric ๐Ÿ„What it captures in the forestLLM hyperparameter ๐Ÿง 
Modularity QFungi form specialized guildsnum_experts (MoE)
Degree distribution ฮณ (scale-free)Few hubs concentrate flowtop_k routing
Small-world coefficient ฯƒLocal signal + global shortcutsnum_heads / attention
Diameter / mean path LHow far carbon travelsnum_layers
Clustering CLocal redundancydropout
Resilience to removalThe network survives failuresregularization / sparsity
Carbon-flow dynamicsPrioritizes by needcurriculum / LR schedule
Preliminary evidence with dashAI

We validated the empirical foundation with dashAI (U. de Chile / CENIA): trained models on mycorrhizal network data to identify critical nodes (hubs), on both clean and noisy data.

F1 1.0
Clean data (3 models)
F1 ~0.68
With realistic noise
Optuna โ‰  better
10 trials did not beat baseline under noise
H3 โœ“
Priors matter with low budgets

The finding: under noisy data and few evaluations, agnostic optimization does not beat the baseline โ€” exactly the regime where a biological prior matters (hypothesis H3 of the paper).

The paper

Myco-Opt: Mycorrhizal-Network-Inspired Priors for LLM-Driven Hyperparameter Optimization
ยท arXiv / NeurIPS (draft v0.2) ยท Open code and data

"Hyperparameter optimization (HPO) and neural architecture search (NAS) remain sample-inefficient despite advances in Bayesian optimization (BO) and LLM-based optimizers such as OPRO. We introduce Myco-Opt, a framework that uses graph-theoretic properties of mycorrhizal networks as biological priors to constrain the search space of LLM-driven optimizers... We hypothesize that biological priors reduce the evaluation budget required to reach a target loss by up to 50%, while yielding structurally distinct, more interpretable configurations."

Interested in the intersection of biology and AI optimization?

We are looking for collaborators to scale the benchmark: integrate real GlobalFungi data, build the OPRO loop with fungal priors, and compare against baselines. Everything is open source.