BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260509T094227EDT-9192Pz9LeD@132.216.98.100
DTSTAMP:20260509T134227Z
DESCRIPTION:Abstract\n\nIn this thesis we study generative models for high 
 dimensional static and temporal data with a design principle of balancing 
 the number of parameters\, network complexity and explainability by determ
 ining the factors that represent the data\, in the meantime\, of maintaini
 ng a unified architecture for different vision tasks.\n\nRestricted Boltzm
 ann Machine (RBM) and other similar networks are difficult to extend beyon
 d a limited spatial extent for the reason that the number of parameters gr
 ows exponentially with the large configuration spaces involved. We propose
  a generalization of a hierarchically undirected model that combines both 
 top-down and bottom-up information propagation for image super-resolution 
 tasks. It aggregates nearby sub-receptive fields to form a larger receptiv
 e field by adding a second hidden layer\, while keeping the number of free
  parameters under control by convolutional weight sharing.\n\nFor temporal
  data\, we focus on the fundamental principle of computer vision\, that is
 \, temporal correlations are the variations between related images which a
 re caused by independent factors - object appearance and motion. The goal 
 is to represent the underlying explanatory factors using decoupling rather
  than keeping them mixed. Once decoupled\, each factor lies in a lower dim
 ensional abstract space. Different computer vision tasks can be conducted 
 more efficiently in corresponding spaces than they are in the original pix
 el space. We present an algorithm that decouples object appearance and loc
 ation to amplitude and phase in static image by using complex factorizatio
 n over orthogonal filter pairs. The filter pairs are learned in an unsuper
 vised manner from multiple consecutive frames. We demonstrate that using t
 his factorization\, object movements are encoded in the phase gradient bet
 ween frames over time by an experiment of optical flow recovery. Test resu
 lts show that small disparity is successfully captured by the factorized p
 hase gradient.\n\nIn separate but related work\, we consider a stochastic 
 mining simulation and put forward a solution using RBM with two-phase lear
 ning. Test results show that this approach offers significant improvements
  to conventional pattern-based algorithms as the RBM is better able to lea
 rn the underlying distribution of the sample data.\n\nWe believe that by c
 onsidering the structural elements of neural networks\, we can gain some i
 nsight into how to develop architectures that can be trained using standar
 d gradient based methods and can tackle more complex problems without grow
 th in complexity.\n
DTSTART:20250919T133000Z
DTEND:20250919T151500Z
LOCATION:Room 603\, McConnell Engineering Building\, CA\, QC\, Montreal\, H
 3A 0E9\, 3480 rue University
SUMMARY:PhD defence of Yanyan Mu – From Foundational Components to Complex 
 Factorizations: Task-Dependent Architectures via Undirected Graphical Mode
 ls
URL:https://www.mcgill.ca/ece/channels/event/phd-defence-yanyan-mu-foundati
 onal-components-complex-factorizations-task-dependent-architectures-367139
END:VEVENT
END:VCALENDAR