Se ofrecen los siguientes trabajos de fin de titulación:
TFG: Comparative Analysis of Single vs. Multi-threaded Transcoding in FFmpeg: Performance Evaluation and Optimization Strategies
OBJETIVE:
This research aims to conduct a comparative analysis between single-threaded and multi-threaded transcoding methodologies within FFmpeg. By evaluating performance metrics such as processing speed, CPU utilization, and memory usage, the study seeks to identify the advantages and limitations of each approach in handling video transcoding tasks. Additionally, the research will assess the quality of experience (QoE) of the transcoded media obtained from both methodologies, examining factors such as visual fidelity and overall user satisfaction.
REQUIREMENTS:
FFmpeg; virtualization or containerization software; Python or Bash; statistical analysis
PERSONA DE CONTACTO:
Alberto del Río – arp@gatv.ssr.upm.es
TFG: Control de Producción de Medios para una Mejor Creación de Contenido
OBJETIVO:
La tesis propuesta tiene como objetivo desarrollar un módulo de control de producción de medios para optimizar los procesos de creación de contenido. El módulo de Producción de Estudio integrará información de diversas fuentes y empleará técnicas innovadoras para producir contenido multimedia que combine de manera fluida y atractiva estos diferentes elementos en un flujo cohesivo.
REQUISITOS:
Python, Herramientas Multimedia (OBS).
PERSONA DE CONTACTO:
Álvaro Llorente – alg@gatv.ssr.upm.es
TFG: Energy Consumption Analysis in Network Probes: Evaluating the Efficiency of Data Transmission between Two Points
OBJETIVE:
This research aims to investigate the energy consumption involved in data transmission between two points using network probes. The study will measure performance metrics such as throughput, latency, and packet loss while focusing on the energy efficiency of the transmission process. The probe, currently implemented using sockets and Python, will be tested to research about the consumption between two end-points. In addition, the research will also explore the impact of different network configurations and protocols on energy efficiency.
REQUIREMENTS:
Python; Networking knowledge (sockets, protocols); Tools for measuring energy consumption; Statistical analysis; Machine learning.
PERSONA DE CONTACTO:
Alberto del Río – arp@gatv.ssr.upm.es
TFG: Benchmarking State-of-the-Art Large Language Models (LLMs): Installation and Performance Comparison
OBJETIVE:
This research aims to install and benchmark several state-of-the-art Large Language Models (LLMs) locally, including Llama, Claude, Gemma, and Mistral. The study will involve researching the requirements for each LLM in terms of hardware, software, and memory. Performance metrics such as inference speed, memory usage, and accuracy will be evaluated and compared across the different models. The goal is to provide a comprehensive comparison of these LLMs to identify their strengths and weaknesses in various application scenarios.
REQUIREMENTS:
Programming skills; Machine Learning; Software.
PERSONA DE CONTACTO:
Alberto del Río – arp@gatv.ssr.upm.es
TFM: Intelligent Model Selection for Efficient Large Language Model Utilization
OBJETIVO:
This research aims to develop an intelligent system that dynamically selects the appropriate model of a Large Language Model (LLM) based on the complexity of the input question. By understanding the input and evaluating the requirements for the response, the system will choose between big models, which offer excellent performance but higher energy costs, and smaller models, which are more efficient for simpler tasks. The goal is to optimize the balance between performance, inference time, and energy consumption.
REQUISITOS:
Programming skills; Machine Learning; Software.
PERSONA DE CONTACTO:
Alberto del Río – arp@gatv.ssr.upm.es
TFG: Automatic Analysis and Conversion of Technical Documentation to Knowledge Graphs
OBJETIVE:
This research aims to develop a tool that automatically analyzes and converts technical documentation into a pivot knowledge graph model using the explicable triple (subject-predicate-object). The study will focus on processing textual parts of the documentation, utilizing advances in Natural Language Processing (NLP) and Large Language Models (LLMs).
REQUIREMENTS:
Strong programming skills (Python); Familiarity with NLP and LLMs; Basic knowledge of knowledge graphs.
PERSONA DE CONTACTO:
Alberto del Río – arp@gatv.ssr.upm.es
TFM: Deep Learning Techniques for Processing Graphical Parts of Technical Documentation
OBJETIVE:
This research aims to develop a function that processes the graphical parts of technical documentation using deep learning and NLP techniques. The study will focus on detecting and recognizing parts (including types and numbers, their role, relationship, and behavior), the operations that can be performed with them, and the order of these operations.
REQUISITOS:
Strong programming skills (Python); Familiarity with NLP and LLMs; Basic knowledge of knowledge graphs.
PERSONA DE CONTACTO:
Alberto del Río – arp@gatv.ssr.upm.es
TFM: Sistema de up-scaling basado en deep-learing para transformar señales de TV de HD a UHD con muy alta calidad, HDR y WCG
Objetivo: El TFM implementará un sistema para realizar up-scaling de señal de vídeo digital HD a UHD, incluyendo cambios en el rango dinámico y en la colorimetría de la imagen, haciendo uso de deep learning. El objetivo final es que dicho sistema de up-scaling ofrezca muy alta calidad, y que funcione en tiempo real. Para ello se cuenta con la colaboración de RTVE y de su infraestructura de computación, dado que RTVE aspira a utilizar dicho sistema de up-scaling para su canal de TV UHD. Este TFM se oferta en el marco de la Cátedra RTVE en la UPM, que será quien proporcione contenidos para los procesos de entrenamiento y aprendizaje de la red, y contará con una ayuda económica en forma de beca, a través del programa Impulsa RTVE.
Requisitos: Conocimientos y experiencia de programación en python y del desarrollo de sistemas de deep learning con imágenes y/o vídeo digital.
Persona de contacto: José Manuel Menéndez – jm.menendez@upm.es