TriProbe traces task separability across model stages to explain failures
TRIPROBE: Probing Task Separability Beyond Classification for XAI
Artificial IntelligenceMachine Learning
Summary
Evaluating how well a machine learning model works usually focuses on its final accuracy, but this leaves unclear why it succeeds or fails. The authors developed TriProbe, a method that looks inside the model at different stages—from raw inputs to features to final predictions—to see how well it separates tasks. TriProbe breaks down multi-task problems into pairs and uses three tests to find where learning problems happen. This helps designers know where to improve data or model architecture. They tested it on a gesture dataset and found hidden issues that standard accuracy misses.
What this means in practice
- •For machine learning engineers: Identify which input features or model components limit task performance to guide targeted improvements in multi-task models.
- •For biomedical signal processing teams: Diagnose failure points in EMG-based gesture recognition systems to optimize data collection and classification stages.
Tested on one dataset.
Authors
Amirhossein Sadough, Freek Hens, Aleksa Bokšan, Mohammad Mahdi Dehshibi, Mahyar Shahsavari
Abstract
Modern evaluation of learning pipelines often reduces to downstream accuracy, leaving open the question of why tasks succeed or fail. TriProbe addresses this gap with a multi-level probing framework for explainable diagnosis of task separability. Rather than treating models as black boxes, TriProbe traces how separability evolves across inputs, learned features, and final classifiers. It decomposes multi-task problems into binary subtasks and applies three complementary probes: a Foundational Probe on input spaces, a Latent Probe on feature representations, and a Final Probe on classifier outputs. Using Maximum Fisher's Discriminant Ratio as a principled separability metric, TriProbe identifies bottlenecks and affected task pairs. Experiments on the Roshambo sEMG benchmark show how TriProbe reveals hidden breakdowns, guiding data collection, validation, and architecture design.