Baseline-Free Policy Optimization Shows Promise for Neural Combinatorial Optimization
Researchers evaluated Group Relative Policy Optimization (GRPO), an algorithm originally developed for large language model alignment, as a replacement for the standard REINFORCE-with-rollout-baseline method used in neural combinatorial optimization (NCO). The standard approach suffers from a structural vulnerability on harder problem instances, where a poor baseline can destabilize training — a collapse observed in experiments on TSP-100 where solution cost degraded from 9.8 to 52.1. GRPO avoids this collapse and achieves solution quality within 2% of a strong established baseline, suggesting it may be a more robust training approach for routing problems.
Neural combinatorial optimization (NCO) uses autoregressive deep learning policies to solve routing problems such as the Travelling Salesman Problem (TSP) and the Capacitated Vehicle Routing Problem (CVRP). The dominant training algorithm, REINFORCE with a rollout baseline, reduces gradient variance by maintaining a periodically updated frozen copy of the policy, but this mechanism can produce noisy gradient estimates on harder instances, potentially destabilizing training. The researchers conducted a controlled comparison of five reinforcement learning algorithms within the RL4CO framework, finding that REINFORCE exhibited a severe training collapse on TSP-100 — solution cost jumping from 9.8 to 52.1 after the warmup phase with no recovery under extended training. GRPO, which eliminates the external baseline by normalizing advantages across groups of sampled trajectories, avoided this collapse entirely. At matched gradient update counts, GRPO achieved solution quality within 2% of POMO, a strong multi-start baseline method. A second alignment-derived algorithm, P3O (a pairwise preference method), was competitive on TSP but showed higher variability on CVRP. The authors conclude that GRPO is a promising baseline-free alternative, especially in settings where baseline-dependent training is fragile.
What's missing
The study benchmarks on TSP and CVRP but does not evaluate generalization to other combinatorial optimization problem types. Computational cost comparisons (wall-clock time, memory) between GRPO and REINFORCE are not reported, which is relevant for practical adoption. The paper also does not assess performance on real-world or large-scale industrial routing instances beyond the benchmark sizes tested.
What different sources said
- arXiv cs.LGCenter
Baseline-Free Policy Optimization for Neural Combinatorial Optimization
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.