Examinando por Materia "Embeddings"

Mostrando 1 - 4 de 4
  • No hay miniatura disponible
    Publicación
    Agente de inteligencia artificial para el apoyo a la primera impresión diagnóstica a partir de descripciones sintomáticas expresadas en lenguaje natural
    (Universidad EAFIT, 2025-11-24) Bertel Morales, Juan Pablo; Jaramillo Múnera, Yomin Estiven
    This thesis proposes the development of an artificial intelligence (AI) agent capable of supporting the generation of an initial diagnostic impression based on symptoms expressed in natural language. The project is grounded in the recognition that medical diagnosis is a complex task prone to errors, particularly when it relies on subjective and unstructured descriptions. To support clinical decision-making, natural language processing and machine learning techniques were applied following the CRISP-DM methodology. The model was trained using the synthetic DDxPlus dataset, which enabled the simulation of clinical scenarios without compromising real patient information. In the process, symptoms were transformed into synthetic anamneses through semantic normalization and subsequently vectorized using various biomedical embedding models. These representations were then used to train a supervised model tasked with associating each narrative with the confirmed diagnosis. As an additional evaluation, a “stress test” was conducted in a simulated environment, in which a healthcare professional interacted directly with the system to assess its ability to interpret real symptomatic descriptions and generate preliminary diagnostic suggestions in a coherent, consistent, and safe.
  • No hay miniatura disponible
    Publicación
    Standardized Approaches for Assessing Metagenomic Contig Binning Performance from Barnes-Hut t-Stochastic Neighbor Embeddings
    (SPRINGER, 2020-01-01) Ceballos J.; Ariza-Jiménez L.; Pinel N.; Universidad EAFIT. Departamento de Ciencias; Bioiversidad, Evolución y Conservación
    The performance of unsupervised methods for metagenomic binning is often assessed using simulated microbial communities. The lack of well-characterized evaluation protocols and approaches to community construction cognizant of biological realities impedes the rigorous assessment and standardization of the binning process. This work attempted to standardize performance evaluation using benchmark communities constructed according to the genome similarity metric Average Amino Acid identity. This approach allowed us to extend and deepen our previous research on the unsupervised binning of metagenomic sequence fragments based on low-dimensional embeddings of pentamer frequency profiles. Experimental results evidenced our method’s potential for the binning of metagenomic contigs to become an alternative to state-of-the-art methods such as MetaCluster 3.0. © 2020, Springer Nature Switzerland AG.
  • No hay miniatura disponible
    Ítem
    Standardized Approaches for Assessing Metagenomic Contig Binning Performance from Barnes-Hut t-Stochastic Neighbor Embeddings
    (SPRINGER, 2020-01-01) Ceballos J.; Ariza-Jiménez L.; Pinel N.; Ceballos J.; Ariza-Jiménez L.; Pinel N.; Universidad EAFIT. Departamento de Ciencias; Ciencias Biológicas y Bioprocesos (CIBIOP)
    The performance of unsupervised methods for metagenomic binning is often assessed using simulated microbial communities. The lack of well-characterized evaluation protocols and approaches to community construction cognizant of biological realities impedes the rigorous assessment and standardization of the binning process. This work attempted to standardize performance evaluation using benchmark communities constructed according to the genome similarity metric Average Amino Acid identity. This approach allowed us to extend and deepen our previous research on the unsupervised binning of metagenomic sequence fragments based on low-dimensional embeddings of pentamer frequency profiles. Experimental results evidenced our method’s potential for the binning of metagenomic contigs to become an alternative to state-of-the-art methods such as MetaCluster 3.0. © 2020, Springer Nature Switzerland AG.
  • No hay miniatura disponible
    Ítem
    Standardized Approaches for Assessing Metagenomic Contig Binning Performance from Barnes-Hut t-Stochastic Neighbor Embeddings
    (SPRINGER, 2020-01-01) Ceballos J.; Ariza-Jiménez L.; Pinel N.; Ceballos J.; Ariza-Jiménez L.; Pinel N.; Universidad EAFIT. Departamento de Ciencias; Modelado Matemático
    The performance of unsupervised methods for metagenomic binning is often assessed using simulated microbial communities. The lack of well-characterized evaluation protocols and approaches to community construction cognizant of biological realities impedes the rigorous assessment and standardization of the binning process. This work attempted to standardize performance evaluation using benchmark communities constructed according to the genome similarity metric Average Amino Acid identity. This approach allowed us to extend and deepen our previous research on the unsupervised binning of metagenomic sequence fragments based on low-dimensional embeddings of pentamer frequency profiles. Experimental results evidenced our method’s potential for the binning of metagenomic contigs to become an alternative to state-of-the-art methods such as MetaCluster 3.0. © 2020, Springer Nature Switzerland AG.