Esci dai Frame

	Web Mining e Retrieval (a.a. 2016/17) Thesis proposal Page
SAG Thesis Proposals and Related Materials
SAG Semantic Analytics Laboratory References for Thesis Proposals. The following list provides material and references to the thesis proposals made available by the SAG Semantic Analytics Laboratory under the coordination of Roberto Basili and Danilo Croce. This list is and always be part of our working and thinking ... in progress. So, please check the list time to time in order to locate new ideas and resources proposed that are possibly of interest for your exam. Thesis 1. Show and Tell in Italian: Study, Definition and Implementation of a Deep Learning architectures for the Automatic Generation of Captions in Italian of raw images. Available Architectures: Existing Neural Network architectures (e.g. implemented within Tensorflow) Dataset: Pascal VOC 2008 / Flickr8k / Flickr30k / MSCOCO datasets Reference papers and resources: Oriol Vinyals, Alexander Toshev, Samy Bengio and Dumitru Erhan Show and Tell: A Neural Image Caption Generator In proceedings of CVPR 2015: 3156-3164 Oriol Vinyals, Alexander Toshev, Samy Bengio and Dumitru Erhan Show and Tell: Lessons learned from the 2015 MSCOCO Image Captioning Challenge In IEEE Transactions on Pattern Analysis and Machine Intelligence. Google Research Blog Thesis 2. Linearizing complex Kernel function for the effective training of high performing Neural Networks. Available Architectures: Nystrom Method implemented in Kelp and Neural Architectures to be evaluated. Dataset: Show&Tell inglese - Community Question Answering Dataset - ImageNet - Human Robot Interaction Corpus Reference papers and resources: Shiliang Sun, Jing Zhao, Jiang Zhu A Review of Nystrom Methods for Large-Scale Machine Learning In Journal Information Fusion archive Volume 26 Issue C, November 2015 Pages 36-48 Danilo Croce, Roberto Basili Large-Scale Kernel-Based Language Learning Through the Ensemble Nystrom Methods. In Advances in Information Retrieval, Volume 9626 of the series Lecture Notes in Computer Science pp 100-112 Thesis 3. Question Answering over the Revealer Architecture Available Architectures: Revealer + (Question Processing) + (Anwer Matching & Re-ranking) Dataset: A dedicated dataset will be made available (e.g., derived among existing TREC datasets) Reference papers and resources: Roberto Basili, Andrea Ciapetti, Danilo Croce, Valeria Marino, Paolo Salvatore, Valerio Storch Enabling Enterprise Semantic Search through Language Technologies: the ProgressIt Experience In Proceedings of the Italian Information Retrieval Workshop, 2014 Antonio Sorgente, Nadia Brancati, Cristina Giannone, Fabio Massimo Zanzotto, Francesco Mele, Roberto Basili Chatting to Personalize and Plan Cultural Itineraries. In Proceedings of UMAP Workshops, 2013 Thesis 4. Attention based Neural Networks for explanation of inductive inferences Available Architectures: To be investigated Dataset: ImageNet/MSCoco or dataset of Reviews (e.g., MovieLENS or from Tripadvisor) Reference papers and resources: M Denil, A Demiraj, N Kalchbrenner, P Blunsom, N de Freitas Modelling, Visualising and Summarising Documents with a Single Convolutional Neural Network In arXiv preprint arXiv:1406.3830 Kelvin Xu, Jimmy Ba, Ryan Kiros, Kyunghyun Cho, Aaron Courville, Ruslan Salakhutdinov, Richard Zemel, Yoshua Bengio Show, Attend and Tell: Neural Image Caption Generation with Visual Attention arXiv preprint arXiv:1502.03044 A brief introduction to Attention in Deep Learning