Overview

Welcome to the APSIPA ASC 2025 Special Session on Scalable and Efficient Signal Processing for Multimodal AI Systems (ESPRESSO). This session will bring together experts and enthusiasts from academia and industry to discuss how we can develop energy-efficient, scalable signal processing techniques for modern AI systems that handle multiple data modalities (such as speech, vision, text, and sensor data). 

As part of APSIPA ASC 2025 – themed “Signal and Information Processing in the Era of Multimodal AI” – our special session aligns with the conference’s focus on cutting-edge multimodal technologies. We warmly invite researchers, engineers, and practitioners to join us in exploring innovative solutions that make multimodal AI more efficient and practical for real-world applications. 

Motivation 

Multimodal AI – which combines data from different sources like audio, images, text, and sensors – holds transformative potential across many domains. From mobile health monitoring and wearable sensing to speech recognition and industrial automation, integrating diverse data streams can lead to richer insights and more powerful AI systems. However, deploying these advanced models outside the lab remains challenging. Key obstacles include managing the complexity of multimodal data, meeting strict low-latency requirements (e.g. real-time responses), and ensuring energy efficiency for AI models running on resource-constrained devices at the edge (such as smartphones or IoT sensors).

Recent progress in AI, especially with Large Language Models (LLMs) and multimodal AI systems, has highlighted the urgent need for scalable and efficient signal processing techniques. As models like multimodal LLMs increasingly fuse text, vision, speech, and other data, they become computationally intensive and difficult to deploy in practical settings. Bridging this gap between state-of-the-art AI and real-world use requires innovative approaches that balance performance, efficiency, and scalability. In other words, we must rethink algorithms and architectures to maintain high accuracy while dramatically reducing computation, memory, and power usage.

Why a special session now? To accelerate progress, this special session provides a dedicated platform for researchers and practitioners to present and discuss the latest innovations in efficient signal and information processing for multimodal AI. Our primary objective is to advance new methodologies for designing scalable, energy-efficient learning models that can be effectively deployed across a range of applications. We place particular emphasis on mobile health, wearable technologies, speech/audio processing, and edge analytics for industrial systems – areas where efficient multimodal AI can be especially impactful. By bringing together experts from diverse disciplines, we hope to foster collaboration and chart a clear roadmap for overcoming these critical challenges.

Potential Topics

Including but not limited to:

• Scalable Learning Models for LLMs and Multimodal LLMs.

• Lightweight AI models under memory and compute constraints for on-device applications.

• Efficient Signal Processing for Wearable Sensing and Mobile Health Monitoring.

• Robust Speech and Audio Processing Systems for Multimodal AI.

• Edge Analytics for Resource-Constrained Devices.

• Low-Latency Learning Models for Real-Time Processing.

• Explainable AI Techniques for Multimodal Systems.

• Energy-Efficient Neural Architectures for Multimodal Integration.

• Transfer Learning and Domain Adaptation for Multimodal Data.

• Data Fusion Techniques for Enhanced Multimodal AI Systems.

• Benchmarks and Evaluation Protocols for Scalable AI Models.

Paper Submission

The Special Session follows the same guidelines as the APSIPA ASC 2025 main conference. When submitting your paper, select “Scalable and Efficient Signal Processing for Multimodal AI Systems” as the subject area. Submitted papers will go through the same review process as the regular papers.

• Submission Deadline of Paper: 11st July 2025

• Notification of Paper Acceptance: 31st July 2025

• Submission Deadline of Camera Ready Paper: 7th August 2025 

Organizers

• Yang Xiao, Fortemedia Singapore (Primary contact) 

• Rohan Kumar Das, Fortemedia Singapore

• Ting Dang, The University of Melbourne

• Eng Siong Chng, Nanyang Technological University