Listed below are the talks to be given by some of our doctoral students during the IPAL full-day meeting, general council and team-building held at Shangri-La Rasa Sentosa.
"Automatic Neuronal Reconstruction" by Sreetama Basu
Neuronal structures are intricately related to their functions. Study of the neuronal structures reveals healthy and pathologic conditions, crucial to understanding how the Brain works. Current advances in microscopy techniques produce huge volumes of data where manual reconstruction and analysis may take several years. The lack of powerful computational tools to automatically reconstruct neuronal arbors has emerged as a major technical bottleneck in neuroscience research. This presentation reviews the existing algorithms and explores the possibility of designing an automatic neuronal reconstruction algorithm based on Marked Point Process.
"Video Crowd Behavior Analysis" by Antoine Fagette
Abstract : Understanding through video analysis the behaviors of a crowd and the situation the observer is facing is a real need that the professional of the domain express even more urgently since CCTV networks are now present and growing faster than ever in every public places. Nowadays, two distinct means -- Video Analysis and Crowd Simulation -- using two different approaches -- Agent-based and Holistic -- are exploited to study crowds. We investigate here the possibility to bind these two means using a holistic approach. Based on works of the past two decades, we decide to explore the resemblance between crowds and fluids.
"A biologically inspired approach to image classification" by Sepehr Jalali
Object recognition in cortex is thought to be mediated by the ventral visual pathway running from primary visual cortex, V1, over extrastriate visual areas V2 and V4 to inferotemporal cortex, IT. Over the last decades, several physiological studies in non-human primates have established a core of basic facts about cortical mechanisms of recognition that seem to be widely accepted and that confirm and refine older data from neuropsychology. A brief summary of this consensus knowledge begins with the ground-breaking work of Hubel and Wiesel. Starting from simple cells in primary visual cortex, V1, with small receptive fields that respond preferably to oriented bars, neurons along the ventral stream show an increase in receptive field size as well as in the complexity of their preferred stimuli. HMAX is a hierarchical computational model of object recognition in cortex proposed by Riesenhuber and Poggio. The standard model simulates the feed-forward path of the visual cortex and has been used to classify animal vs. non-animal images and clip images first and is similar to Neocognitron in using both simple and complex cells. This model is used to find a trade-off between invariance and selectivity. We explore this model in details and provide modifications to fit the model better for classification tasks.
"Semantic Web for AAL: Design & Deployment - Lessons Learned" by Thibaut Tiberghien
Robust solutions for ambient assisted living are numerous, yet predominantly specific in their scope of usability. This presentation describes the contribution of the service oriented architecture combined with semantic web technologies in order to build more versatile solutions — a step towards adaptable context-aware engines and simplified deployments. Our design and deployment work in hindsight, we highlight some requirements for semantic web tools that would help with real-life deployment of semantically driven assistive technologies. We also describe our current service platform developed as an integrated solution for AAL.
"Incorporation of prior-knowledge into learning with SVMs" by Antoine Veillard
SVMs with the general purpose RBF kernel are widely considered as state-of-the-art supervised learning algorithms due to their effectiveness and versatility. However, in practice, SVMs often require more training data than readily available. Prior-knowledge may be available to compensate this shortcoming provided such knowledge can be effectively passed on to SVMs. We propose a method for the incorporation of prior-knowledge via an adaptation of the standard RBF kernel. Our practical and computationally simple approach allows prior-knowledge in a variety of forms ranging from regions of the input space as crisp or fuzzy sets to pseudo-periodicity. We show that this method is effective and that the amount of required training data can be largely decreased, opening the way for new usages of SVMs.