We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overriding in the fully-connected layers we employed a recently-developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.

/pdf/imagenet-classification-with-deep-convolutional-neural-1190x9a6b9.pdf

ImageNet Classification with Deep Convolutional Neural Networks

We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0%, respectively, which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overfitting in the fully connected layers we employed a recently developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.

/pdf/imagenet-classification-with-deep-convolutional-neural-3kv7rezkq1.pdf

ImageNet classification with deep convolutional neural networks

As the Netflix Prize competition has demonstrated, matrix factorization models are superior to classic nearest neighbor techniques for producing product recommendations, allowing the incorporation of additional information such as implicit feedback, temporal effects, and confidence levels

/pdf/matrix-factorization-techniques-for-recommender-systems-1ygvnzk1c8.pdf

Matrix Factorization Techniques for Recommender Systems

Today's Web-enabled deluge of electronic data calls for automated methods of data analysis. Machine learning provides these, developing methods that can automatically detect patterns in data and then use the uncovered patterns to predict future data. This textbook offers a comprehensive and self-contained introduction to the field of machine learning, based on a unified, probabilistic approach. The coverage combines breadth and depth, offering necessary background material on such topics as probability, optimization, and linear algebra as well as discussion of recent developments in the field, including conditional random fields, L1 regularization, and deep learning. The book is written in an informal, accessible style, complete with pseudo-code for the most important algorithms. All topics are copiously illustrated with color images and worked examples drawn from such application domains as biology, text processing, computer vision, and robotics. Rather than providing a cookbook of different heuristic methods, the book stresses a principled model-based approach, often using the language of graphical models to specify models in a concise and intuitive way. Almost all the models described have been implemented in a MATLAB software package--PMTK (probabilistic modeling toolkit)--that is freely available online. The book is suitable for upper-level undergraduates with an introductory-level college math background and beginning graduate students.

Machine Learning : A Probabilistic Perspective

We propose a unified neural network architecture and learning algorithm that can be applied to various natural language processing tasks including part-of-speech tagging, chunking, named entity recognition, and semantic role labeling. This versatility is achieved by trying to avoid task-specific engineering and therefore disregarding a lot of prior knowledge. Instead of exploiting man-made input features carefully optimized for each task, our system learns internal representations on the basis of vast amounts of mostly unlabeled training data. This work is then used as a basis for building a freely available tagging system with good performance and minimal computational requirements.

/pdf/natural-language-processing-almost-from-scratch-osj5vem4f3.pdf

Natural Language Processing (Almost) from Scratch

The collaborative filtering (CF) approach to recommenders has recently enjoyed much interest and progress. The fact that it played a central role within the recently completed Netflix competition has contributed to its popularity. This chapter surveys the recent progress in the field. Matrix factorization techniques, which became a first choice for implementing CF, are described together with recent innovations. We also describe several extensions that bring competitive accuracy into neighborhood methods, which used to dominate the field. The chapter demonstrates how to utilize temporal models and implicit feedback to extend models accuracy. In passing, we include detailed descriptions of some the central methods developed for tackling the challenge of the Netflix Prize competition.

/pdf/advances-in-collaborative-filtering-kahzf23jul.pdf

Advances in Collaborative Filtering

This article outlines the overall strategy and summarizes a few key innovations of the team that won the first Netflix progress prize.

/pdf/lessons-from-the-netflix-prize-challenge-18kr24czgp.pdf

Lessons from the Netflix prize challenge

Advance knowledge of which files in the next release of a large software system are most likely to contain the largest numbers of faults can be a very valuable asset. To accomplish this, a negative binomial regression model has been developed and used to predict the expected number of faults in each file of the next release of a system. The predictions are based on the code of the file in the current release, and fault and modification history of the file from previous releases. The model has been applied to two large industrial systems, one with a history of 17 consecutive quarterly releases over 4 years, and the other with nine releases over 2 years. The predictions were quite accurate: for each release of the two systems, the 20 percent of the files with the highest predicted number of faults contained between 71 percent and 92 percent of the faults that were actually detected, with the overall average being 83 percent. The same model was also used to predict which files of the first system were likely to have the highest fault densities (faults per KLOC). In this case, the 20 percent of the files with the highest predicted fault densities contained an average of 62 percent of the system's detected faults. However, the identified files contained a much smaller percentage of the code mass than the files selected to maximize the numbers of faults. The model was also used to make predictions from a much smaller input set that only contained fault data from integration testing and later. The prediction was again very accurate, identifying files that contained from 71 percent to 93 percent of the faults, with the average being 84 percent. Finally, a highly simplified version of the predictor selected files containing, on average, 73 percent and 74 percent of the faults for the two systems.

Predicting the location and number of faults in large software systems

Recommender systems based on collaborative filtering predict user preferences for products or services by learning past user-item relationships. A predominant approach to collaborative filtering is neighborhood based ("k-nearest neighbors"), where a user-item preference rating is interpolated from ratings of similar items and/or users. We enhance the neighborhood-based approach leading to substantial improvement of prediction accuracy, without a meaningful increase in running time. First, we remove certain so-called "global effects" from the data to make the ratings more comparable, thereby improving interpolation accuracy. Second, we show how to simultaneously derive interpolation weights for all nearest neighbors, unlike previous approaches where each weight is computed separately. By globally solving a suitable optimization problem, this simultaneous interpolation accounts for the many interactions between neighbors leading to improved accuracy. Our method is very fast in practice, generating a prediction in about 0.2 milliseconds. Importantly, it does not require training many parameters or a lengthy preprocessing, making it very practical for large scale applications. Finally, we show how to apply these methods to the perceivably much slower user-oriented approach. To this end, we suggest a novel scheme for low dimensional embedding of the users. We evaluate these methods on the netflix dataset, where they deliver significantly better results than the commercial netflix cinematch recommender system.

http://brettb.net/project/papers/2007%20Scalable%20collaborative%20filtering%20with%20jointly%20derived%20neighborhood%20interpolation%20weights.pdf

Robert M. Bell

Papers

Matrix Factorization Techniques for Recommender Systems

Advances in Collaborative Filtering

Lessons from the Netflix prize challenge

Predicting the location and number of faults in large software systems

Scalable Collaborative Filtering with Jointly Derived Neighborhood Interpolation Weights