Search

Many recent advancements in Computer Vision are attributed to large datasets. Open-source software packages for Machine Learning and inexpensive commodity hardware have reduced the barrier of entry for exploring novel approaches at scale. It is possible to train models over millions of examples within a few days. Although large-scale datasets exist for image understanding, such as ImageNet, there are no comparable size video classification datasets. In this paper, we introduce YouTube-8M, the largest multi-label video classification dataset, composed of ~8 million videos (500K hours of video), annotated with a vocabulary of 4800 visual entities. To get the videos and their labels, we used a YouTube video annotation system, which labels videos with their main topics. While the labels are machine-generated, they have high-precision and are derived from a variety of human-based signals including metadata and query click signals. We filtered the video labels (Knowledge Graph entities) using both automated and manual curation strategies, including asking human raters if the labels are visually recognizable. Then, we decoded each video at one-frame-per-second, and used a Deep CNN pre-trained on ImageNet to extract the hidden representation immediately prior to the classification layer. Finally, we compressed the frame features and make both the features and video-level labels available for download. We trained various (modest) classification models on the dataset, evaluated them using popular evaluation metrics, and report them as baselines. Despite the size of the dataset, some of our models train to convergence in less than a day on a single machine using TensorFlow. We plan to release code for training a TensorFlow model and for computing metrics., Comment: 10 pages

Current layouts for alphabetic input on mobile phone keypads are very inefficient. We propose a genetic algorithm (GA) to find a suitable keypad layout for each user, based on their personal text history. It incorporates codes for frequent multigrams, which may be directly input. This greatly reduces the average number of strokes required for typing. We optimize for two-handed use, the left thumb covering the leftmost rows and vice versa. The GA mini- mizes the number of strokes, consecutive use of the same key, and consecutive use of the same hand. Using these criteria, the algorithm re-arranges the 26 alphabetic characters, plus 14 additional multigrams, on the 10-key pad. We demonstrate that this arrangement can generate a more effective layout, especially for SMS-style messages. Substantial savings are verified by both computational analysis and human evaluation., Comment: 18 pages (including header pages), 3 figures, 8 tables

A huge number of academic papers are coming out from a lot of conferences and journals these days. In these circumstances, most researchers rely on key-based search or browsing through proceedings of top conferences and journals to find their related work. To ease this difficulty, we propose a Personalized Academic Research Paper Recommendation System, which recommends related articles, for each researcher, that may be interesting to her/him. In this paper, we first introduce our web crawler to retrieve research papers from the web. Then, we define similarity between two research papers based on the text similarity between them. Finally, we propose our recommender system developed using collaborative filtering methods. Our evaluation results demonstrate that our system recommends good quality research papers.

Matrix approximation is a common tool in machine learning for building accurate prediction models for recommendation systems, text mining, and computer vision. A prevalent assumption in constructing matrix approximations is that the partially observed matrix is of low-rank. We propose a new matrix approximation model where we assume instead that the matrix is only locally of low-rank, leading to a representation of the observed matrix as a weighted sum of low-rank matrices. We analyze the accuracy of the proposed local low-rank modeling. Our experiments show improvements in prediction accuracy in recommendation tasks., Comment: 3 pages, 2 figures, Workshop submission to the First International Conference on Learning Representations (ICLR)

Collaborative filtering is a rapidly advancing research area. Every year several new techniques are proposed and yet it is not clear which of the techniques work best and under what conditions. In this paper we conduct a study comparing several collaborative filtering techniques -- both classic and recent state-of-the-art -- in a variety of experimental contexts. Specifically, we report conclusions controlling for number of items, number of users, sparsity level, performance criteria, and computational complexity. Our conclusions identify what algorithms work well and in what conditions, and contribute to both industrial deployment collaborative filtering algorithms and to the research community., Comment: 27 pages, 12 figures