Pasar al contenido principal

ESCALABILIDAD PARA INTELIGENCIA ARTIFICIAL EN EL EDGE MEDIANTE TECNOLOGIAS SOSTENIBLES

PID2024-160996OB-I00

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal para la Investigación y el Desarrollo Experimental
Subprograma Subprograma Estatal de Generación de Conocimiento Científico-Técnico y Desarrollo Experimental
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2024
Plan Estatal 2024-2027 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario UNIVERSITAT POLITECNICA DE CATALUNYA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 3
Encontrada(s) 1 página(s)

RUA. Repositorio Institucional de la Universidad de Alicante
Publicación Otros (Other)

Motivando la sostenibilidad ambiental y el uso responsable de recursos mediante Aprendizaje Profundo en los estudios de Inteligencia Artificial

  • Berral García, Josep Lluís
  • Torres Viñals, Jordi
Los estudios en Inteligencia Artificial (IA) han traído un nuevo tipo de alumnado interesado en la algoritmia y datos, pero desconectado de loa nuevos retos relacionados con el impacto ambiental de la IA y la computación de altas prestaciones. Transmitir tales conocimientos requiere una metodología docente que fomente la motivación y el interés hacia el coste en recursos y efecto del uso de herramientas basadas en IA. Mediante el uso de t´ecnicas de aprendizaje basado en ejemplos, basado en retos, y actividades que fomentan la interacción y colaboración, se busca conseguir que el alumnado vea de primera mano la necesidad de entender el coste ambiental y en recursos de la IA. Además, se busca conseguir así que se forme como profesional con mentalidad crítica y consciente del balance entre altas prestaciones y sostenibilidad ambiental de los recursos de computación. A partir de unas primeras experiencias, hemos observado como tales técnicas han ayudado a que el alumnado tome consciencia y interés, comprobando una mejora en la implicación y consideración respecto a estos temas, además de recibir una retroalimentación positiva.
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Adaptive prompt evolution for continual learning in diabetic retinopathy detection

  • Bravo Rocca, Gusseppe
  • Liu, Peini
  • Guitart Fernández, Jordi
  • Dholakia, Ajay
  • Ellison, David
  • Carrillo Larco, Rodrigo M.
We introduce Adaptive Prompt Evolution (APE), a framework that autonomously optimizes text prompts for continual learning (CL) in medical imaging applications. Building upon task-agnostic domain-incremental learning, APE addresses static prompt engineering limitations through intelligent feedback loops that evolve templates and medical descriptions using iterative performance assessment. Our approach replaces manual prompt crafting with a local medical language model (Med-Gemma 4B) that generates and refines prompt candidates based on zero-shot clustering performance, maintaining complete offline operation for healthcare compliance. APE demonstrates strong efficiency, requiring only 50 evaluations to achieve F1-score of 0.9229, compared to exhaustive search requiring 110 evaluations for F1-score of 0.8921, representing 2.2X computational efficiency with 3.45% performance improvement. Evaluated on diabetic retinopathy detection across three increasingly complex tasks, we integrate APE with established CL strategies including Elastic Weight Consolidation (EWC), Gradient Episodic Memory (GEM), and Learning Without Forgetting (LwF). On experiments across Multi-Layer Perceptron (MLP), Residual, and Attention architectures, APE outperforms baselines across all strategies. The framework reduces privacy exposure during experience replay by storing only embeddings rather than raw images, while demonstrating robust generalization performance across tasks with minimal degradation within the studied diabetic retinopathy setting., Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature. This work was partially supported by Lenovo and Intel® as part of the Lenovo AI Innovators University Research program, by the Spanish Ministry of Science (MICINN), the Research State Agency (AEI) and European Regional Development Funds (ERDF/FEDER) under grant agreement PID2024-160996OB-I00, MICIU/AEI/10.13039/501100011033, and by the Generalitat de Catalunya under contract 2021-SGR-00478., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Data-driven optimization of GPU efficiency for distributed LLM-adapter serving

  • Agulló López, Ferran
  • Oliveras Torra, Joan
  • Wang, Chen
  • Gutiérrez Torre, Alberto
  • Tardieu, Olivier
  • Youssef, Alaa
  • Torres Viñals, Jordi
  • Berral García, Josep Lluís
Large Language Model (LLM) adapters enable low-cost model specialization, but introduce complex caching and scheduling challenges in distributed serving systems where hundreds of adapters must be hosted concurrently. While prior work has largely focused on latency and throughput optimization, minimizing GPU resource requirements through near-peak utilization remains largely underexplored. This paper presents a data-driven pipeline that, for a given workload, computes an adapter placement that serves the workload with the minimum number of GPUs while avoiding request starvation and GPU memory errors. To that end, the approach identifies the maximum feasible throughput attainable on each GPU by leveraging accurate performance predictions learned from real serving behavior. The proposed pipeline integrates three components: (i) a Digital Twin (DT) tailored to LLM-adapter serving, (ii) a distilled machine learning (ML) model trained on DT-generated data, and (iii) a greedy placement algorithm that exploits ML-based performance estimates to maximize GPU efficiency. The DT emulates real system dynamics with high fidelity, achieving below 5% throughput estimation error while executing up to 90×faster than full LLM benchmarking across both predictable and unpredictable workloads. The learned ML models further accelerate performance estimation with marginal accuracy degradation, enabling scalable optimization. Experimental results demonstrate that the pipeline substantially improves GPU efficiency, reducing the number of GPUs required to sustain target workloads by 60% on average across the evaluated scenarios. Beyond GPU efficiency, the pipeline can be adapted to alternative objectives, such as latency minimization, highlighting its versatility for future large-scale LLM serving infrastructures., This work has been partially financed by the EU-HORIZON MSCA programme under grant agreement GA.101086248. Also, it has been partially financed by Generalitat de Catalunya (AGAUR) under grant agreement 2021-SGR-00478, by Severo Ochoa Center of Excellence CEX-2021-001148-S-20-3, and by the Spanish Ministry of Science (MICINN), the Research State Agency (AEI) and European Regional Development Funds (ERDF/FEDER) under grant agreement PID2024-160996OB-I00, MICIU/AEI/10.13039/501100011033/FEDER, UE., Peer Reviewed, Postprint (published version)