Image recognition and classification is one of the primary challenges of the machine learning community. Recent advances in learning systems, coupled with hardware developments have enabled general object recognition systems to be learned on home computers with graphics processing units. Presented is a Deep Belief Network engineered using NVIDIA's CUDA programming language for general object recognition tasks.

A neurone model (the FORMON) is proposed which provides a mathematical explanation for a range of psychological phenomena and has potential in Artificial Intelligence applications. A general definition of organisation in terms of entropy and information is formulated. The concept of microcodes is introduced to describe the physical nature of organisation. Spatio-temporal pattern acquisition and processing functions attributable to individual neurones are reviewed. The criterion for self-organisation in a neurone is determined as the maximisation of mutual organisation. A feedback control system is proposed to satisfy this criterion and provide an integrated long-term memory of spatio-temporal pattern. This pattern acquisition system is shown to be applicable to dendritic pattern recognition and axonal pattern generation. Provision is also made for adaptation, short-term memory and operant learning. An electro-chemical model of transmission and processing of neural signals is outlined to provide the pattern acquisition functions of the Formon model. A transverse magnetic mode of electrotonic propagation is postulated in addition to the transverse electromagnetic mode. Configurations of the Formon are categorised in terms of possible pattern processing functions. Connective architectures are proposed as self-organising models of acquisitive semantic and syntactic networks.

abstract: Real-world environments are characterized by non-stationary and continuously evolving data. Learning a classification model on this data would require a framework that is able to adapt itself to newer circumstances. Under such circumstances, transfer learning has come to be a dependable methodology for improving classification performance with reduced training costs and without the need for explicit relearning from scratch. In this thesis, a novel instance transfer technique that adapts a "Cost-sensitive" variation of AdaBoost is presented. The method capitalizes on the theoretical and functional properties of AdaBoost to selectively reuse outdated training instances obtained from a "source" domain to effectively classify unseen instances occurring in a different, but related "target" domain. The algorithm is evaluated on real-world classification problems namely accelerometer based 3D gesture recognition, smart home activity recognition and text categorization. The performance on these datasets is analyzed and evaluated against popular boosting-based instance transfer techniques. In addition, supporting empirical studies, that investigate some of the less explored bottlenecks of boosting based instance transfer methods, are presented, to understand the suitability and effectiveness of this form of knowledge transfer. / Dissertation/Thesis / M.S. Computer Science 2011

Indiana University-Purdue University Indianapolis (IUPUI) / Pedestrian Automatic Emergency Braking (PAEB) system for avoiding/mitigating pedestrian crashes have been equipped on some passenger vehicles. At present, there are many e orts for the development of common standard for the performance evaluation of PAEB. The Transportation Active Safety Institute (TASI) at Indiana University-Purdue University-Indianapolis has been studying the problems and ad- dressing the concerns related to the establishment of such a standard with support from Toyota Collaborative Safety Research Center (CSRC). One of the important components in the PAEB evaluation is the development of standard testing facili- ties at night, in which 70% pedestrian crash social costs occurs [1]. The test facility should include representative low-illuminance environment to enable the examination of sensing and control functions of di erent PAEB systems. This thesis work focuses on modeling low-illuminance driving environment and describes an approach to recon- struct the lighting conditions. The goal of this research is to characterize and model light sources at a potential collision case at low-illuminance environment and deter- mine possible recreation of such environment for PAEB evaluation. This research is conducted in ve steps. The rst step is to identify lighting components that ap- pear frequently on a low-illuminance environment that a ect the performance of the PAEB. The identi ed lighting components include ambient light, same side/opposite side light poles, opposite side car headlight. Next step is to collect all potential pedes- trian collision cases at night with GPS coordinate information from TASI 110 CAR naturalistic driving study video database. Thirdly, since ambient lighting is relatively random and lack of a certain pattern, ambient light intensity for each potential col- lision case is de ned and processed as the average value of a region of interest on all video frames in this case. Fourth step is to classify interested light sources from the selected videos. The temporal pro le method, which compressing region of interest in video data (x,y,t) to image data (x,y), is introduced to scan certain prede ned region on the video. Due to the fact that light sources (except ambient light) impose distinct light patterns on the road, image patterns corresponding to speci c light sources can be recognized and classi ed. All light sources obtained are stamped with GPS coordinates and time information which are provided in corresponding data les along with the video. Lastly, by grouping all light source information of each repre- sentative street category, representative light description of each street category can be generated. Such light description can be used for lighting construction of PAEB test facility.