Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Shapley Features for Robust Signal Prediction in Tactile Internet

Mohammad Ali Vahedifar, Qi Zhang

TL;DR

This work tackles the challenge of robust, low-latency haptic signal prediction in the Tactile Internet by introducing a two-stage framework where a Gaussian Process (GP) oracle guides a ResNet-based neural network (NN). A Jensen-Shannon Divergence loss aligns the NN's predicted distribution with the GP oracle, while Shapley Feature Values (SFV) perform offline feature selection to identify the most informative inputs, reducing computation without sacrificing accuracy. Empirically, GP+SFV achieves 95.72% accuracy (outperforming LeFo by 11.1%), with substantial inference-time reductions (GP refit every 10 samples and per-sample inference around 2.2 ms). The approach demonstrates a practical, high-accuracy, and efficient solution for bidirectional haptic communication in TI, leveraging offline analysis to support real-time operations and extending SFV as a general-purpose accelerator for TI signal processing.

Abstract

The Tactile Internet (TI) requires ultra-low latency and reliable haptic signal transmission, yet packet loss and delay remain unresolved challenges. We present a novel prediction framework that integrates Gaussian Processes (GP) with a ResNet-based Neural Network, where GP acts as an oracle to recover signals lost or heavily delayed. To further optimize performance, we introduce Shapley Feature Values (SFV), a principled feature selection mechanism that isolates the most informative inputs for prediction. This GP+SFV framework achieves 95.72% accuracy, surpassing the state-of-the-art LeFo method by 11.1%, while simultaneously relaxing TI's rigid delay constraints. Beyond accuracy, SFV operates as a modular accelerator: when paired with LeFo, it reduces inference time by 27%, and when paired with GP, by 72%. These results establish GP+SFV as both a high-accuracy and high-efficiency solution, paving the way for practical and reliable haptic communications in TI systems.

Shapley Features for Robust Signal Prediction in Tactile Internet

TL;DR

This work tackles the challenge of robust, low-latency haptic signal prediction in the Tactile Internet by introducing a two-stage framework where a Gaussian Process (GP) oracle guides a ResNet-based neural network (NN). A Jensen-Shannon Divergence loss aligns the NN's predicted distribution with the GP oracle, while Shapley Feature Values (SFV) perform offline feature selection to identify the most informative inputs, reducing computation without sacrificing accuracy. Empirically, GP+SFV achieves 95.72% accuracy (outperforming LeFo by 11.1%), with substantial inference-time reductions (GP refit every 10 samples and per-sample inference around 2.2 ms). The approach demonstrates a practical, high-accuracy, and efficient solution for bidirectional haptic communication in TI, leveraging offline analysis to support real-time operations and extending SFV as a general-purpose accelerator for TI signal processing.

Abstract

The Tactile Internet (TI) requires ultra-low latency and reliable haptic signal transmission, yet packet loss and delay remain unresolved challenges. We present a novel prediction framework that integrates Gaussian Processes (GP) with a ResNet-based Neural Network, where GP acts as an oracle to recover signals lost or heavily delayed. To further optimize performance, we introduce Shapley Feature Values (SFV), a principled feature selection mechanism that isolates the most informative inputs for prediction. This GP+SFV framework achieves 95.72% accuracy, surpassing the state-of-the-art LeFo method by 11.1%, while simultaneously relaxing TI's rigid delay constraints. Beyond accuracy, SFV operates as a modular accelerator: when paired with LeFo, it reduces inference time by 27%, and when paired with GP, by 72%. These results establish GP+SFV as both a high-accuracy and high-efficiency solution, paving the way for practical and reliable haptic communications in TI systems.

Paper Structure

This paper contains 8 sections, 1 theorem, 11 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. System Model
  4. Gaussian Process
  5. Neural Network Training with JSD Loss
  6. Shapley Feature Value
  7. Results & Discussion
  8. Conclusion

Key Result

Theorem 1

Any Feature valuation $\phi_a$ satisfying Axioms 1 and 2 must have the form: If and only if we calculate Eq. Eq:ShapleyValue for valuation, then Axioms 1 and 2 are satisfied.

Figures (3)

  • Figure 1: Overview of two-stage predictive framework for TI.
  • Figure 2: Prediction accuracy and standard deviation for predicting the next 10 samples with ResNet training with GP+SFV.
  • Figure 3: (a) Shapley Feature Value analysis for GP+SFV on ResNet for Drag Max Stiffness Y Dataset. (b) Error prediction of the next 10 samples with ResNet. Solid lines: Human, dashed lines: Robot.

Theorems & Definitions (1)

  • Theorem 1: Shapley Feature Value