Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ILASH: A Predictive Neural Architecture Search Framework for Multi-Task Applications

Md Hafizur Rahman, Md Mashfiq Rizvee, Sumaiya Shomaji, Prabuddha Chakraborty

TL;DR

This work tackles the energy and carbon challenges of neural architecture search and edge inference by introducing ILASH, a layer-sharing multi-task architecture, and ILASH-NAS, an AI-guided NAS framework. ILASH combines strategic layer reuse across tasks with branching to support additional tasks, while ILASH-NAS employs heuristic and predictive search to build efficient models under device constraints, including an energy-aware Goodness metric. Experiments on UTKFace, MTFL, CelebA, and Taskonomy show substantial gains, including up to 16x energy/CO2 reductions during NAS and training and up to 3x improvements during inference, with predictive ILASH-NAS (ILASH-Pred) achieving large speedups (e.g., ~38x faster NAS) and energy savings over heuristic methods. The results demonstrate practical impact for deploying multitask AI on resource-constrained edge devices, and lay groundwork for broader AI-guided NAS with energy-awareness.

Abstract

Artificial intelligence (AI) is widely used in various fields including healthcare, autonomous vehicles, robotics, traffic monitoring, and agriculture. Many modern AI applications in these fields are multi-tasking in nature (i.e. perform multiple analysis on same data) and are deployed on resource-constrained edge devices requiring the AI models to be efficient across different metrics such as power, frame rate, and size. For these specific use-cases, in this work, we propose a new paradigm of neural network architecture (ILASH) that leverages a layer sharing concept for minimizing power utilization, increasing frame rate, and reducing model size. Additionally, we propose a novel neural network architecture search framework (ILASH-NAS) for efficient construction of these neural network models for a given set of tasks and device constraints. The proposed NAS framework utilizes a data-driven intelligent approach to make the search efficient in terms of energy, time, and CO2 emission. We perform extensive evaluations of the proposed layer shared architecture paradigm (ILASH) and the ILASH-NAS framework using four open-source datasets (UTKFace, MTFL, CelebA, and Taskonomy). We compare ILASH-NAS with AutoKeras and observe significant improvement in terms of both the generated model performance and neural search efficiency with up to 16x less energy utilization, CO2 emission, and training/search time.

ILASH: A Predictive Neural Architecture Search Framework for Multi-Task Applications

TL;DR

This work tackles the energy and carbon challenges of neural architecture search and edge inference by introducing ILASH, a layer-sharing multi-task architecture, and ILASH-NAS, an AI-guided NAS framework. ILASH combines strategic layer reuse across tasks with branching to support additional tasks, while ILASH-NAS employs heuristic and predictive search to build efficient models under device constraints, including an energy-aware Goodness metric. Experiments on UTKFace, MTFL, CelebA, and Taskonomy show substantial gains, including up to 16x energy/CO2 reductions during NAS and training and up to 3x improvements during inference, with predictive ILASH-NAS (ILASH-Pred) achieving large speedups (e.g., ~38x faster NAS) and energy savings over heuristic methods. The results demonstrate practical impact for deploying multitask AI on resource-constrained edge devices, and lay groundwork for broader AI-guided NAS with energy-awareness.

Abstract

Artificial intelligence (AI) is widely used in various fields including healthcare, autonomous vehicles, robotics, traffic monitoring, and agriculture. Many modern AI applications in these fields are multi-tasking in nature (i.e. perform multiple analysis on same data) and are deployed on resource-constrained edge devices requiring the AI models to be efficient across different metrics such as power, frame rate, and size. For these specific use-cases, in this work, we propose a new paradigm of neural network architecture (ILASH) that leverages a layer sharing concept for minimizing power utilization, increasing frame rate, and reducing model size. Additionally, we propose a novel neural network architecture search framework (ILASH-NAS) for efficient construction of these neural network models for a given set of tasks and device constraints. The proposed NAS framework utilizes a data-driven intelligent approach to make the search efficient in terms of energy, time, and CO2 emission. We perform extensive evaluations of the proposed layer shared architecture paradigm (ILASH) and the ILASH-NAS framework using four open-source datasets (UTKFace, MTFL, CelebA, and Taskonomy). We compare ILASH-NAS with AutoKeras and observe significant improvement in terms of both the generated model performance and neural search efficiency with up to 16x less energy utilization, CO2 emission, and training/search time.

Paper Structure

This paper contains 19 sections, 1 equation, 3 figures, 4 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Methodology
  3. ILASH: Layer Shared Architecture
  4. Heuristic Based ILASH-NAS
  5. Predictive ILASH-NAS
  6. Experimental Setup and Data Description
  7. Experimental Setup: Auto-ILASH
  8. Experimental Setup: Edge Devices
  9. Experimental Setup: Task Models
  10. Data Description
  11. UTKFace
  12. MTFL
  13. CelebA
  14. Taskonomy
  15. Experimental Results and Analysis
  16. ...and 4 more sections

Figures (3)

  • Figure 1: Overview of the proposed ILASH architecture and the neural architecture search algorithm (ILASH-NAS).
  • Figure 2: Inferencing setup using target edge devices.
  • Figure 3: UTKFace, MTFL, and CelebA Datasets: Neural search efficiency and model accuracy comparison between ILASH and other state-of-the-art multitask NAS algorithms.