SmartMixed: Two-Phase Training Strategy Enables Neural Networks to Learn Optimal Per-Neuron Activation Functions
Researchers introduced SmartMixed, a training method that allows individual neurons in neural networks to learn and select their own optimal activation functions from a pool of candidates rather than using a single fixed function across all neurons. The approach uses a two-phase strategy: first, neurons adaptively select activation functions through a differentiable mechanism, then selections are fixed for efficient inference. Testing on MNIST showed that neurons in different layers prefer different activation functions, suggesting functional diversity benefits network performance.
SmartMixed addresses a fundamental design choice in neural networks by enabling per-neuron activation function learning instead of relying on uniform, fixed functions across architectures. The method operates in two phases: during training, neurons use a differentiable hard mixture mechanism to select from six candidate activation functions (ReLU, Sigmoid, Tanh, Leaky ReLU, ELU, SELU), and after learning converges, selections are frozen to maintain computational efficiency at inference time. Evaluation on MNIST with feedforward networks revealed that neurons exhibit layer-dependent preferences for specific activation functions, indicating that heterogeneous activation choices may improve network performance. The approach maintains computational efficiency through vectorized operations while allowing continued training after the selection phase. This work suggests that the common practice of using identical activation functions across all neurons may be suboptimal, and that learned heterogeneous activation strategies could enhance neural network design.
What's missing
The study is limited to MNIST evaluation with feedforward networks; generalization to larger datasets (CIFAR-10, ImageNet), modern architectures (CNNs, Transformers), and comparison with other adaptive activation function methods are not addressed. The paper does not discuss computational overhead during the first training phase or provide theoretical justification for why per-neuron selection should improve performance.
What different sources said
- arXiv cs.AICenter
SmartMixed: A Two-Phase Training Strategy for Adaptive Activation Function Learning in Neural Networks
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.