BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:Asia/Kolkata BEGIN:STANDARD DTSTART:19451014T230000 TZOFFSETFROM:+0630 TZOFFSETTO:+0530 TZNAME:IST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20230917T073100Z UID:B2997D98-F36D-4773-B46B-3A1FCDFC0DA9 DTSTART;TZID=Asia/Kolkata:20230915T151500 DTEND;TZID=Asia/Kolkata:20230915T171500 DESCRIPTION:Throughout the event\, attendees witnessed the demystification of the world of DNN as its true potential in data analysis was unlocked. L earning with meaningful constraints and the implementation of sparse codin g techniques left a profound impact on feature extraction. By imposing lim itations and emphasizing sparsity\, distinctive and sparse discriminative features were extracted from the data. These features became essential com ponents of the original sets of objects and were visually represented as s parse basis vectors.\nFurthermore\, the integration of sparse basis functi ons\, such as receptive fields or filters\, enhanced transparency by effec tively overlaying and reconstructing them with minimal reconstruction erro r. This approach ensured a clear and comprehensive understanding of the un derlying concepts\, marking a significant milestone in the field of data r epresentation.\n\nTechniques discussed are\n(1) Nonnegative Matrix Factori zation that reduces the number of basis vectors and allows for extraction of latent features that are additive and hence interpretable for humans.\n (2) A classic EBP architecture can also be trained under the constraints o f nonnegativity and sparseness. The resulting classifiers allow for identi fication of parts of the objects encoded as receptive fields developed by weights of hidden neurons. The results are illustrated with MNIST handwrit ten digits classifiers and Reuters-21578 text categorization.\n(3) A const rained learning of sparse non-negative weights in auto-encoders also allow s for discovery of additive latent factors. Our experiments with MNIST dat asets compare the auto-encoder accuracy for various training conditions. T hey indicate enhanced interpretability and insights through identification of parts of complex input objects traded-off for a small reduction of rec ognition accuracy or classification error.\n\nSpeaker(s): Jacek M. Zurada\ , \n\nAgenda: \nThe event\, offered attendees a unique opportunity to expe rience the future of data\nrepresentation. Deep neural networks (DNN) had long grappled with issues related to\ntransparency and complex mappings in discriminative data representation. However\,\nduring the event\, experts unveiled cutting-edge techniques that addressed these\nchallenges\, ultim ately bidding farewell to the complexities caused by cancellations of\npos itive and negative terms. These breakthroughs ushered in a new era of clar ity and\nefficiency in the computation process.\n\nRoom: Seminar Hall\, Bl dg: Department of CSE\, JAWAHARLAL NEHRU TECHNOLOGICAL UNIVERSITY HYDERABA D\, KUKATPALLY HOUSING BOARD COLONY\, KUKATPALLY\, Hyderabad\, Andhra Prad esh\, India\, 500085\, Virtual: https://events.vtools.ieee.org/m/373231 LOCATION:Room: Seminar Hall\, Bldg: Department of CSE\, JAWAHARLAL NEHRU TE CHNOLOGICAL UNIVERSITY HYDERABAD\, KUKATPALLY HOUSING BOARD COLONY\, KUKAT PALLY\, Hyderabad\, Andhra Pradesh\, India\, 500085\, Virtual: https://eve nts.vtools.ieee.org/m/373231 ORGANIZER: SEQUENCE:20 SUMMARY:Inside the Black Box: Deep Learning with Sparse Coding\, Additive F eatures and Nonnegative Matrix Decompositions URL;VALUE=URI:https://events.vtools.ieee.org/m/373231 X-ALT-DESC:Description: <br /><p>Throughout the event\, attendees witnessed the demystification of the world of DNN as its true potential in data ana lysis was unlocked. Learning with meaningful constraints and the implement ation of sparse coding techniques left a profound impact on feature extrac tion. By imposing limitations and emphasizing sparsity\, distinctive and s parse discriminative features were extracted from the data. These features became essential components of the original sets of objects and were visu ally represented as sparse basis vectors.<br />Furthermore\, the integrati on of sparse basis functions\, such as receptive fields or filters\, enhan ced transparency by effectively overlaying and reconstructing them with mi nimal reconstruction error. This approach ensured a clear and comprehensiv e understanding of the underlying concepts\, marking a significant milesto ne in the field of data representation.</p>\n<p>Techniques discussed are<b r />(1) Nonnegative Matrix Factorization that reduces the number of basis vectors and allows for extraction of latent features that are additive and hence interpretable for humans.<br />(2) A classic EBP architecture can a lso be trained under the constraints of nonnegativity and sparseness. The resulting classifiers allow for identification of parts of the objects enc oded as receptive fields developed by weights of hidden neurons. The resul ts are illustrated with MNIST handwritten digits classifiers and Reuters-2 1578 text categorization.<br />(3) A constrained learning of sparse non-ne gative weights in auto-encoders also allows for discovery of additive late nt factors. Our experiments with MNIST datasets compare the auto-encoder a ccuracy for various training conditions. They indicate enhanced interpreta bility and insights through identification of parts of complex input objec ts traded-off for a small reduction of recognition accuracy or classificat ion error.</p><br /><br />Agenda: <br /><p>The event\, offered attendees a unique opportunity to experience the future of data&nbsp\;<br />represent ation. Deep neural networks (DNN) had long grappled with issues related to &nbsp\;<br />transparency and complex mappings in discriminative data repr esentation. However\,&nbsp\;<br />during the event\, experts unveiled cutt ing-edge techniques that addressed these&nbsp\;<br />challenges\, ultimate ly bidding farewell to the complexities caused by cancellations of&nbsp\;< br />positive and negative terms. These breakthroughs ushered in a new era of clarity and&nbsp\;<br />efficiency in the computation process.</p> END:VEVENT END:VCALENDAR