Multi-Armed Bandit Algorithms Improve Structured Pruning of Neural Networks
Researchers have developed a structured neural network pruning framework that uses multi-armed bandit (MAB) algorithms to selectively remove entire neurons, reducing model size while maintaining or improving performance. The approach treats each candidate neuron as a 'bandit arm,' temporarily masking it to measure its impact on loss before deciding whether to permanently remove it. The work offers a computationally practical alternative to unstructured pruning, which is harder to exploit in standard hardware implementations.
A new preprint posted to arXiv proposes using multi-armed bandit (MAB) algorithms to perform structured pruning of deep neural networks by removing complete neurons rather than individual weights. The framework evaluates several MAB policies — including Epsilon-Greedy, UCB1, Thompson Sampling, Softmax, Hedge-style multiplicative weights, and EXP3 — by treating each neuron as an arm whose reward reflects the safety of its removal. Experiments span tabular classification, tabular regression, and deep learning benchmarks across image, text, and reasoning tasks. Statistical analysis using the Friedman test and Nemenyi post-hoc test found significant differences between policies: UCB1 achieved the highest mean rank on both classification and regression tasks, while UCB1 and Thompson Sampling led on deep learning benchmarks. Several MAB policies outperformed not only the unpruned baseline but also magnitude-based and greedy activation-variation pruning methods. The authors argue that structured pruning via MAB algorithms is both effective and computationally practical, as removing whole neurons yields hardware-friendly sparse models.
What's missing
The study does not report wall-clock training or pruning times relative to baseline methods, making it difficult to assess the practical computational overhead of the MAB evaluation loop. It is also unclear how sensitive results are to the mini-batch size used during the reward estimation step, or how the framework scales to very large modern architectures (e.g., transformer models with billions of parameters). The paper is a preprint and has not yet undergone peer review.
What different sources said
- arXiv cs.AICenter
Structured Neuron Pruning in Deep Neural Networks Using Multi-Armed Bandits
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.