BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Departamento de Computación - ECPv6.15.18//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Departamento de Computación
X-ORIGINAL-URL:https://www.dc.uba.ar
X-WR-CALDESC:Eventos para Departamento de Computación
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:America/Sao_Paulo
BEGIN:STANDARD
TZOFFSETFROM:-0300
TZOFFSETTO:-0300
TZNAME:-03
DTSTART:20220101T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=America/Sao_Paulo:20230302T140000
DTEND;TZID=America/Sao_Paulo:20230302T150000
DTSTAMP:20260408T182015
CREATED:20230224T155651Z
LAST-MODIFIED:20230224T155651Z
UID:8126-1677765600-1677769200@www.dc.uba.ar
SUMMARY:Charla Data Quality for Deep Learning
DESCRIPTION:Los invitamos a todos a la charla «Data Quality for Deep Learning» dictada por el Prof. Dr. Saúl Calderón Ramirez. \nDía y horario: jueves 2 de marzo\, 14hs.\nLugar: aula a confirmar en el pabellón 0+i. \nAbstract:\nDeep learning models usually need extensive amounts of data\, and these data have to be labeled\, becoming a concern when dealing with real-world applications. It is known that labeling a dataset is a costly task in time\, money\, and resource-wise. Different methods exploit small labelled datasets and other types of data with  less costly labelling schemes  (Data augmentation\, self-supervised learning\, semi supervised learning\, etc.). For instance\, Semi-supervised Learning Model (SSLM) uses labeled and unlabeled datasets to train a model\, improving the overall performance of the models when labeled datasets are small. The unlabeled datasets may include out-of-distribution data with respect to the labeled data\, which may affect the model’s accuracy and future predictions. We introduce the importance of data quality metrics\, especially when considering that the future of Deep learning models targets real-world applications such as healthcare. Concepts such as data quality metrics has been normally applied in structured data\, however\, it can also be applied in unstructured data (datasets used to train deep learning models\, in different types of learning settings. \nSaúl Calderón Ramirez es Ph. D. en Cs. de la Computación  (Universidad De Montfort\, Reino Unido)\, y Magister Scientae en Ingeniería Eléctrica con énfasis en sistemas digitales (Universidad de Costa Rica). Saúl es Profesor en el Instituto Tecnológico de Costa Rica y coordina el PAttern Recognition and MAchine Learning Group (PARMA-Group).
URL:https://www.dc.uba.ar/event/charla-data-quality-for-deep-learning/
LOCATION:Aula a confirmar
CATEGORIES:Agenda
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Sao_Paulo:20230302T160000
DTEND;TZID=America/Sao_Paulo:20230302T170000
DTSTAMP:20260408T182015
CREATED:20230224T155545Z
LAST-MODIFIED:20230228T130157Z
UID:8124-1677772800-1677776400@www.dc.uba.ar
SUMMARY:Charla Computing with Infinite objects: the Gray code case
DESCRIPTION:Están todos invitados a esta charla en el Departamento de Computación: \nProfesor Dieter Spreen\,  University of Siegen\, Germany\, Department of Mathematics \nTítulo :  Computing with Infinite objects: the Gray code case \nDía:    Jueves 2 de  marzo 2023\nHora: 16 hs\nLugar  Cero más Infinito\, Ciudad Universitaria\, aula 1115 \nAbstract: In theoretical studies on exact computations with real numbers\, but as well in applications\, the signed digit representation of the reals is mostly used. This is an extension of the binary expansion of the reals which is very redundant: Every real number is represented by infinitely many infinite strings over the alphabet {-1\, 0\, 1}. Infinite Gray code\, on the other hand\, has no redundancy: it is a one-to-one representation of the reals. Nevertheless\, there are computable transformations between the two representations. For transforming Gray code into signed digit code\, however\, the algorithm must make use of nondeterministic concurrent computations. \nIn the talk a representation-free logic-based approach is presented in which the representation are expressed as predicates S and G on the reals. The realizers of statements S(x) and G(x) are exactly the respective codes of x. From a proof of the inclusion of S in G one can extract a correct program transforming signed digit into Gray code. For a proof of the converse inclusion\, however\, one needs to extend the logic by new connectives which are realized by concurrent computations. As will be shown\, the approach can be extended from the number case to the case of nonempty subsets of the reals.
URL:https://www.dc.uba.ar/event/charla-computing-with-infinite-objects-the-gray-code-case/
LOCATION:Aula 1115
CATEGORIES:Agenda
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/Sao_Paulo:20230327T140000
DTEND;TZID=America/Sao_Paulo:20230327T150000
DTSTAMP:20260408T182015
CREATED:20230321T122434Z
LAST-MODIFIED:20230327T145402Z
UID:8148-1679925600-1679929200@www.dc.uba.ar
SUMMARY:Defensa Tesis Licenciatura Nicolas Mastropasqua
DESCRIPTION:Título: Estudio de redes profundas livianas con aprendizaje basado en distribuciones aplicadas al reconocimiento de expresiones faciales.\nDirector: Daniel Acevedo.\nJurados: Pablo Negri y Juan Manuel Pérez. \nResumen:\nHoy en día\, la búsqueda de soluciones ‘lightweight’ que logren resultados comparables a modelos de Deep Learning robustos ha recibido particular atención debido a su implementación factible en dispositivos móviles. Uno de los problemas que podrían aprovechar esta cualidad es el de Facial Expression Recognition (FER). Considerando que la mayoría de los datasets de expresiones faciales suelen estar anotados con emociones categóricas cuando en realidad la mayoría de las expresiones exhibidas en escenarios ‘in the wild’ ocurren como combinaciones o composición de emociones básicas\, se puede hacer uso de Label Distibution Learning (LDL) como estrategia para el entrenamiento.\nEn este trabajo se abordará el problema de FER a través de redes neuoronales livianas usando LDL en modelos de Deep Learning livianos. Bajo el supuesto de que las imágenes de expresiones faciales deberían tener una distribución de emoción similar a la de su vecindad en un espacio de etiquetas auxiliares adecuado\, como aquel determinado por la tarea de Action Unit Recognition\, se puede aprovechar la información de las distribuciones e incorporarla como parte la función de pérdida.\nConcretamente\, se estudiarán en profundidad dos arquitecturas ‘lightweight’ del Estado del arte\, EfficientFace y CERN\, y se analizará el impacto de distintos acercamientos para implementar LDL considerando datasets ‘in the wild’ como RAF-DB\, CAER-S\, FER+ y AffectNet.
URL:https://www.dc.uba.ar/event/defensa-tesis-licenciatura-nicolas-mastropasqua/
LOCATION:Aula 1203
CATEGORIES:Agenda
END:VEVENT
END:VCALENDAR