Researchers Propose Parameter-Efficient Adapter for Multimodal Learning with Tabular and Image Data
A new machine learning framework called TI-Adapter enables efficient fine-tuning of models that combine tabular data and images by using targeted adapters rather than full model retraining. The approach freezes pretrained encoders and adds lightweight adapter layers at specific points in the network architecture. Testing on 20 datasets shows the method matches or exceeds full fine-tuning performance while requiring substantially fewer trainable parameters.
Researchers have introduced TI-Adapter, a parameter-efficient fine-tuning framework designed for multimodal learning tasks that integrate structured tabular data with visual information. The method addresses a key challenge in machine learning: pretrained models provide strong representations but full fine-tuning is computationally expensive, while keeping models frozen limits task-specific adaptation. TI-Adapter solves this by freezing the pretrained tabular encoder and strategically inserting lightweight adapters—at the embedding level for tabular data and at both embedding and bottleneck levels for images. Comprehensive experiments across 20 tabular-image datasets demonstrate that TI-Adapter achieves competitive or superior predictive performance compared to full fine-tuning while using significantly fewer trainable parameters. Ablation studies confirm that the placement of adapters is critical for balancing model performance with practical computational efficiency.
What different sources said
- arXiv cs.LGCenter
Parameter-Efficient Adapter Tuning for Tabular-Image Multimodal Learning
Related
Genetic Drift, Not Selection, Drives Rapid Feather Color Evolution in Island Bird Radiation
A new study of an island bird radiation found that rapid evolution of feather coloration is driven primarily by genetic drift in small populations rather than sexual or ecological selection. The research integrated whole-genome data with detailed plumage measurements across complete species sampling to test whether signaling trait evolution correlates with speciation rates. The findings suggest that neutral demographic processes play a central role in generating phenotypic diversity during island radiations, challenging assumptions about the mechanisms driving rapid evolution.
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.