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

In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.

Very Deep Convolutional Networks for Large-Scale Image Recognition

The explosion of image data on the Internet has the potential to foster more sophisticated and robust models and algorithms to index, retrieve, organize and interact with images and multimedia data. But exactly how such data can be harnessed and organized remains a critical problem. We introduce here a new database called “ImageNet”, a large-scale ontology of images built upon the backbone of the WordNet structure. ImageNet aims to populate the majority of the 80,000 synsets of WordNet with an average of 500-1000 clean and full resolution images. This will result in tens of millions of annotated images organized by the semantic hierarchy of WordNet. This paper offers a detailed analysis of ImageNet in its current state: 12 subtrees with 5247 synsets and 3.2 million images in total. We show that ImageNet is much larger in scale and diversity and much more accurate than the current image datasets. Constructing such a large-scale database is a challenging task. We describe the data collection scheme with Amazon Mechanical Turk. Lastly, we illustrate the usefulness of ImageNet through three simple applications in object recognition, image classification and automatic object clustering. We hope that the scale, accuracy, diversity and hierarchical structure of ImageNet can offer unparalleled opportunities to researchers in the computer vision community and beyond.

/pdf/imagenet-a-large-scale-hierarchical-image-database-1e8f663s1h.pdf

ImageNet: A large-scale hierarchical image database

We propose a deep convolutional neural network architecture codenamed Inception that achieves the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. By a carefully crafted design, we increased the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC14 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.

/pdf/going-deeper-with-convolutions-1yobw2o2ds.pdf

Going deeper with convolutions

A new trend in email spam is the emergence of image spam. Although current anti-spam technologies are quite successful in filtering text-based spam emails, the new image spams are substantially more difficult to detect, as they employ a variety of image creation and randomization algorithms. Spam image creation algorithms are designed to defeat well-known vision algorithms such as optical character recognition (OCR) algorithms whereas randomization techniques ensure the uniqueness of each image. We observe that image spam is often sent in batches that consist of visually similar images that differ only due to the application of randomization algorithms. Based on this observation, we propose an image spam detection system that uses near-duplicate detection to detect spam images. We rely on traditional anti-spam methods to detect a subset of spam images and then use multiple image spam filters to detect all the spam images that “look” like the spam caught by traditional methods. We have implemented a prototype system to achieve high detection rate while having a less than 0.001% false positive rate.

https://www.cs.princeton.edu/cass/papers/spam_ceas07.pdf

Filtering Image Spam with Near-Duplicate Detection.

There has been much research on checkpointing algorithms for parallel and distributed systems; but surprisingly few implementations for uniprocessors, multiprocessors, and distributed systems, and none at all for multicomputers. We discuss ickp, our consistent checkpointer for the Intel iPSC/860, which is the first general-purpose checkpointer for a multicomputer. It is a checkpointing library that may be invoked asynchronously from the host processor, at a periodic interval, or by a library call. It implements three consistent checkpointing algorithms, two optimizations to reduce checkpoint time and overhead, and recovery. >

ickp: a consistent checkpointer for multicomputers

A disk-based archival storage system including a storage unit configured to store archival data, the storage unit including at least one spindle of disks configured to magnetically store archival data, an interconnect; and a control unit configured to process requests over the interconnect to either archive to or retrieve data from the storage unit. In one embodiment, the system includes a plurality of the storage units, each including at least one spindle of disks. The control unit controls the storage unit(s) in a master-slave relationship. Specifically the control unit is capable of issuing commands to selectively cause the storage unit(s) to shut down or power up, enter a running mode or a standby mode, cause the spindle of disk(s) to either spin up or spin down, and to perform a data integrity check of all the archival data stored in the storage system. In various other embodiments, the control unit runs algorithms that expand the lifetime and longevity of the disk spindles, optimizes power consumption, and performs data migration in the event a data integrity check identifies correctable errors.

Archival data storage system and method

In today's multicomputers, software overhead dominates the message-passing latency cost. We designed two multicomputer network interfaces that significantly reduce this overhead. Both support virtual-memory-mapped communication, allowing user processes to communicate without expensive buffer management and without making system calls across the protection boundary separating user processes from the operating system kernel. Here we compare the two interfaces and discuss the performance trade-offs between them. >

Virtual-memory-mapped network interfaces

A variety of performance-enhancing techniques, such as striping, mirroring, and rotational data replication, exist in the disk array literature. Given a fixed budget of disks, one must intelligently choose what combination of these techniques to employ. In this paper, we present a way of designing disk arrays that can flexibly and systematically reduce seek and rotational delay in a balanced manner. We give analytical models that can guide an array designer towards optimal configurations by considering both disk and workload characteristics. We have implemented a prototype disk array that incorporates the configuration models. In the process, we have also developed a robust disk head position prediction mechanism without any hardware support. The resulting prototype demonstrates the effectiveness of the configuration models.

https://www.usenix.org/publications/library/proceedings/osdi2000/full_papers/yugum/yugum.pdf

Kai Li

Papers

Filtering Image Spam with Near-Duplicate Detection.

ickp: a consistent checkpointer for multicomputers

Archival data storage system and method

Virtual-memory-mapped network interfaces

Trading capacity for performance in a disk array