Researchers Develop Implicit Neural Representations to Learn Policies from Unlabeled Behavioral Data
Researchers have introduced 'Behavioral INR,' a self-supervised model that applies implicit neural representations (INRs) — a technique from computer vision — to identify and distinguish individual behavioral policies from unlabeled, mixed behavioral data. The work addresses a common challenge in robotics, gaming, and autonomous systems where datasets contain heterogeneous behaviors from multiple agents but lack policy annotations. The approach could improve policy identification in real-world settings where labeling is impractical and behaviors are difficult to distinguish.
Presented at the ICML 2026 Structured Probabilistic Inference & Generative Modeling Workshop, Behavioral INR reframes policy representation as a function mapping states to actions, analogous to how vision INRs map pixel coordinates to color values. An episode-level latent variable modulates this function via FiLM conditioning layers, enabling a generative prior over policies and unsupervised inference of policy identity. The model naturally handles variable episode lengths and different sampling granularities, mirroring how vision INRs accommodate varying image resolutions. The authors also introduce a new taxonomy of out-of-distribution (OOD) shifts specific to behavioral data — along state-distribution and action-distribution axes — which are not captured by conventional OOD benchmarks focused solely on new agents or environments. Evaluations span synthetic Gaussian random field data, MuJoCo locomotion demonstrations, real-world chess games, Formula 1 racing telemetry, robotics datasets, and a Seek-Avoid task. Behavioral INR most consistently outperforms baselines in continuous, high-dimensional state-action settings, particularly when episodes are long, policies are numerous, or OOD conditions reduce the utility of simple marginal statistics. Amortized history encoders remain competitive in simpler, symbolic, or low-dimensional settings where policy identity can be recovered from repetition or basic action statistics.
What's missing
The generalization of the OOD taxonomy to domains beyond those tested remains an open question, as does scalability to very large or highly multimodal behavioral datasets.
What different sources said
- arXiv cs.AICenter
Implicit Neural Representations of Individual Behavior
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.