Pol Mareno Comellas/Ian Simpson
Tue 14 Apr 2015, 11:00 - 12:00
Informatics Forum (IF-G.07)

If you have a question about this talk, please contact: Mary-Clare Mackay (mmackay3)

Pol Mareno Comellas

Title: "Vision-as-inverse-graphics for Detailed Scene Understanding"

For tasks such as autonomous robot behaviour, it is important to go beyond object recognition and detection to more detailed descriptions
of the objects in a scene such as their shape, pose, and appearance. I will present the vision-as-inverse-graphics approach to extract these
descriptions. The key idea is to use the known generative process of images (e.g. how light interacts  with objects surfaces to give rise to images)
in order to recover the scene structure. In addition to this generative model, bottom-up recognition models can be used for a more efficient solution
to the vision-as-inverse-graphics problem. I will motivate the combination of both generative and recognition models, and show how we can make
use of synthetic, computer-graphics generated images to  train the recognition models and show how different factors in images (e.g. occlusion and
background clutter) affect their performance.

Ian Simpson

Title:   "Network Analyses Reveal Novel Aspects of ALS Pathogenesis"


Amyotrophic Lateral Sclerosis (ALS) is a fatal neurodegenerative disease characterized by selective loss of motor neurons, muscle atrophy
and paralysis. Mutations in the human VAMP-associated protein B (hVAPB) cause a heterogeneous group of motor neuron diseases including
ALS8. Despite extensive research, the molecular mechanisms underlying ALS pathogenesis remain largely unknown. Genetic screens for key
interactors of hVAPB activity in the intact nervous system, however, represent a fundamental approach towards understanding the in-vivo function
of hVAPB and its role in ALS pathogenesis. Targeted expression of the disease-causing allele leads to neurodegeneration and progressive decline
in motor performance when expressed in the adult Drosophila, eye or in its entire nervous system, respectively. By using these two phenotypic
readouts, we carried out a systematic survey of the Drosophila genome to identify modifiers of hVAPB-induced neurotoxicity. Modifiers cluster
in a diverse array of biological functions including processes and genes that have been previously linked to hVAPB function, such as proteolysis
and vesicular trafficking. In addition to established mechanisms, the screen identified endocytic trafficking and genes controlling proliferation
and apoptosis as potent modifiers of ALS8-mediated defects. Surprisingly, the list of modifiers was mostly enriched for proteins linked to lipid
droplet biogenesis and dynamics. Computational analysis reveals that most modifiers can be linked into a complex network of interacting genes,
and that the human genes homologous to the Drosophila modifiers can be assembled into an interacting network largely overlapping with that in
flies. Identity markers of the endocytic process were also found to abnormally accumulate in ALS patients, further supporting the relevance of
the fly data for human biology. Collectively, these results not only lead to a better understanding of hVAPB function but also point to potentially
relevant targets for therapeutic intervention.