Research map / Architectures and efficiency / RNNs and normalization
Training Recurrent Neural Networks with Inherent Missing Data for Wearable Device Applications (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Opportunistic Teacher-Forcing, which trains RNNs with BPTT by teacher-forcing when observations exist and running autonomously without learning otherwise, handling 80% missing data without imputation on the Lorenz 63 system.
Problem
Training RNNs on wearable physiological time series with large gaps and missing values, where conventional imputation during pre-processing can introduce bias.
Approach
Switch between teacher-forcing and autonomous mode (without learning) based on data availability during BPTT, rather than using scheduled sampling's predefined probability schedule that assumes fully observed data.
Key results
- Opportunistic Teacher-Forcing yields autonomous reconstructions of the Lorenz attractor comparable to Scheduled Sampling under 80% missing data (qualitative delay embeddings).
- Training completed at 9,864 epochs for Opportunistic Teacher-Forcing versus 11,046 epochs for Scheduled Sampling.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Opportunistic Teacher-Forcing instead of Scheduled Sampling | MT-RNN · Lorenz 63, 80% missing data | attractor reconstruction | No clear effect |
Limitations
- Evaluated only on the surrogate Lorenz 63 system, not real wearable data.
- Missing data simulated as uniform random omission with a fixed probability; real missingness patterns may differ.
- Evaluation is qualitative, with no quantitative error metrics reported.
Experimental setup
Datasets: Lorenz 63 (simulated)
Benchmarks: Lorenz 63 attractor reconstruction
Metrics: qualitative delay embedding comparison
Baselines: Scheduled Sampling
Builds on: Scheduled Sampling Multi-Timescale RNN
Closest papers by meaning
- Scaling Wearable Foundation Models (ICLR 2025 · similarity 0.85)
- Scalable Hierarchical Deep Neural Network for Time Series Analysis in Wearable Sensor-based Human Activity Recognition (AAAI 2025 · similarity 0.85)
- Beyond Sensor Data: Foundation Models of Behavioral Data from Wearables Improve Health Predictions (ICML 2025 · similarity 0.84)
- VCR: Learning Valid Contextual Representation for Incomplete Wearable Signals (NeurIPS 2026 · similarity 0.82)
- Large-scale Training of Foundation Models for Wearable Biosignals (ICLR 2024 · similarity 0.82)
- RelCon: Relative Contrastive Learning for a Motion Foundation Model for Wearable Data (ICLR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.