BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Interaction Learning with Neural Cellular Automata
DTSTART:20240705T150000
DTEND:20240705T170000
DTSTAMP:20260510T131855Z
UID:f51200caa4b6ff0c6c296fb25ccc6fee5ad174e9202f74767bd57029
CATEGORIES:Conferences - Seminars
DESCRIPTION:Yitao Xu\nEDIC candidacy exam\nExam president: Prof. Caglar Gu
 lcehre\nThesis advisor: Prof. Sabine Süsstrunk\nCo-examiner: Prof. Mathie
 u Salzmann\n\nAbstract\nNeural Cellular Automata (NCA) models the global c
 ell representation via local cell interactions. Cells in NCA are represent
 ed by real vectors\, allowing various interaction modeling techniques to b
 e adopted. Existing works demonstrate that common operations in deep learn
 ing\, such as convolution on 2D images\, message passing on graphs\, and s
 elf-attention operation can be applied to model cell interactions in NCA. 
 While promising\, the application scenarios of current NCA are mainly on e
 xemplar-based training\, discriminative tasks on small datasets\, and low-
 resolution image generation. The limited size of models and datasets hinde
 rs further real-world applications of NCA on large-scale problems.\n\nIn t
 his thesis\, we will first formulate the evolution paradigm of NCA by intr
 oducing the synergistic combination of cell Interaction and Adaptation in 
 NCA evolution. The focus will be on the Interaction stage. We will introdu
 ce the most commonly used Interaction modeling method\, i.e.\, 2D convolut
 ion. Then\, we will elaborate on the modification of 2D convolution to all
 ow more flexible cell Interaction learning\, which results in message pass
 ing on graph data. Finally\, a cell-dependent interaction scheme is introd
 uced\, where each cell decides its unique interaction strategy via self-at
 tention.\n\nBased on the analysis of NCA paradigm\, we will propose the po
 tential practical application scenarios on those three interaction modelin
 g methods\, and elaborate on the 2D convolution scheme as the first step o
 f extending NCA into real-world applications.\n\nBackground papers\ncoming
  soon
LOCATION:BC 410 https://plan.epfl.ch/?room==BC%20410
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR