Time-Series & Sensors

Learning Using Privileged Information for FRDA Assessment

Learning Using Privileged Information for Objective Assessment of Friedreich’s Ataxia

Training with information you will not have later: genetic and history data guide learning, but only sensor data is needed at test time.

The problem

The best severity predictions use genetic and clinical details that are often unavailable during routine assessment.

The approach

A Learning Using Privileged Information framework reconstructs disease duration and genetic repeat length from cup-sensor data during training, comparing four variants with cross-validated tuning.

The outcome

Pearson r = 0.817 from sensor data alone, beating the sensor-only baseline (0.798).

In detail

Explored Learning Using Privileged Information to improve objective Friedreich ataxia assessment from AIM-C sensor data. Genetic and clinical information such as GAA1 repeat length and disease duration is available to the model during training, but inference can be performed using IMU signals alone. This design targets stronger predictive performance without requiring sensitive or difficult-to-obtain clinical variables during routine deployment.

Built with

• Machine Learning
• IMU Analytics
• Privileged Information

More in Time-Series & Sensors

• Probabilistic Deep Ensembles for FRDA Severity Assessment
• Cross-Device Domain Adaptation for FRDA Assessment
• AIM-M: Multiview IMU Assessment of Friedreich Ataxia

Next project: Forecasting Loss of Ambulation in Friedreich Ataxia

By Kanishka Ranaweera, PhD, Machine Learning Engineer in Melbourne, Victoria, Australia. hello@kanishkabuilt.me