Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

/pdf/deep-residual-learning-for-image-recognition-b75jg61qrq.pdf

Deep Residual Learning for Image Recognition

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

Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers---8x deeper than VGG nets but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. 
The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

Deep learning is a form of machine learning that enables computers to learn from experience and understand the world in terms of a hierarchy of concepts. Because the computer gathers knowledge from experience, there is no need for a human computer operator to formally specify all the knowledge that the computer needs. The hierarchy of concepts allows the computer to learn complicated concepts by building them out of simpler ones; a graph of these hierarchies would be many layers deep. This book introduces a broad range of topics in deep learning. The text offers mathematical and conceptual background, covering relevant concepts in linear algebra, probability theory and information theory, numerical computation, and machine learning. It describes deep learning techniques used by practitioners in industry, including deep feedforward networks, regularization, optimization algorithms, convolutional networks, sequence modeling, and practical methodology; and it surveys such applications as natural language processing, speech recognition, computer vision, online recommendation systems, bioinformatics, and videogames. Finally, the book offers research perspectives, covering such theoretical topics as linear factor models, autoencoders, representation learning, structured probabilistic models, Monte Carlo methods, the partition function, approximate inference, and deep generative models. Deep Learning can be used by undergraduate or graduate students planning careers in either industry or research, and by software engineers who want to begin using deep learning in their products or platforms. A website offers supplementary material for both readers and instructors.

Deep Learning

While microblog readers have a wide variety of reactions to the content they see, studies have tended to focus on extremes such as retweeting and unfollowing. To understand the broad continuum of reactions in-between, which are typically not shared publicly, we designed a website that collected the first large corpus of follower ratings on Twitter updates. Using our dataset of over 43,000 voluntary ratings, we find that nearly 36% of the rated tweets are worth reading, 25% are not, and 39% are middling. These results suggest that users tolerate a large amount of less-desired content in their feeds. We find that users value information sharing and random thoughts above me-oriented or presence updates. We also offer insight into evolving social norms, such as lack of context and misuse of @mentions and hashtags. We discuss implications for emerging practice and tool design.

http://people.csail.mit.edu/msbernst/papers/wgat-cscw2012.pdf

Who gives a tweet?: evaluating microblog content value

Massive online classes are global and diverse. How can we harness this diversity to improve engagement and learning? Currently, though enrollments are high, students' interactions with each other are minimal: most are alone together. This isolation is particularly disappointing given that a global community is a major draw of online classes. This paper illustrates the potential of leveraging geographic diversity in massive online classes. We connect students from around the world through small-group video discussions. Our peer discussion system, Talkabout, has connected over 5,000 students in fourteen online classes. Three studies with 2,670 students from two classes found that globally diverse discussions boost student performance and engagement: the more geographically diverse the discussion group, the better the students performed on later quizzes. Through this work, we challenge the view that online classes are useful only when in-person classes are unavailable. Instead, we demonstrate how diverse online classrooms can create benefits that are largely unavailable in a traditional classroom.

/pdf/talkabout-making-distance-matter-with-small-groups-in-3v3qistsol.pdf

Talkabout: Making Distance Matter with Small Groups in Massive Classes

Web search engines now offer more than ranked results. Queries on topics like weather, definitions, and movies may return inline results called answers that can resolve a searcher's information need without any additional interaction. Despite the usefulness of answers, they are limited to popular needs because each answer type is manually authored. To extend the reach of answers to thousands of new information needs, we introduce Tail Answers: a large collection of direct answers that are unpopular individually, but together address a large proportion of search traffic. These answers cover long-tail needs such as the average body temperature for a dog, substitutes for molasses, and the keyboard shortcut for a right-click. We introduce a combination of search log mining and paid crowdsourcing techniques to create Tail Answers. A user study with 361 participants suggests that Tail Answers significantly improved users' subjective ratings of search quality and their ability to solve needs without clicking through to a result. Our findings suggest that search engines can be extended to directly respond to a large new class of queries.

/pdf/direct-answers-for-search-queries-in-the-long-tail-2oieoo16ao.pdf

Direct answers for search queries in the long tail

To lower the threshold to participation in crowdsourcing, we present twitch crowdsourcing: crowdsourcing via quick contributions that can be completed in one or two seconds. We introduce Twitch, a mobile phone application that asks users to make a micro-contribution each time they unlock their phone. Twitch takes advantage of the common habit of turning to the mobile phone in spare moments. Twitch crowdsourcing activities span goals such as authoring a census of local human activity, rating stock photos, and extracting structured data from Wikipedia pages. We report a field deployment of Twitch where 82 users made 11,240 crowdsourcing contributions as they used their phone in the course of everyday life. The median Twitch activity took just 1.6 seconds, incurring no statistically distinguishable costs to unlock speed or cognitive load compared to a standard slide-to-unlock interface.

/pdf/twitch-crowdsourcing-crowd-contributions-in-short-bursts-of-6z431i3p3h.pdf

Twitch crowdsourcing: crowd contributions in short bursts of time

Errors and biases are earning algorithms increasingly malignant reputations in society. A central challenge is that algorithms must bridge the gap between high-level policy and on-the-ground decisions, making inferences in novel situations where the policy or training data do not readily apply. In this paper, we draw on the theory of street-level bureaucracies, how human bureaucrats such as police and judges interpret policy to make on-the-ground decisions. We present by analogy a theory of street-level algorithms, the algorithms that bridge the gaps between policy and decisions about people in a socio-technical system. We argue that unlike street-level bureaucrats, who reflexively refine their decision criteria as they reason through a novel situation, street-level algorithms at best refine their criteria only after the decision is made. This loop-and-a-half delay results in illogical decisions when handling new or extenuating circumstances. This theory suggests designs for street-level algorithms that draw on historical design patterns for street-level bureaucracies, including mechanisms for self-policing and recourse in the case of error.

https://dl.acm.org/doi/pdf/10.1145/3290605.3300760

Michael S. Bernstein

Papers

Who gives a tweet?: evaluating microblog content value

Talkabout: Making Distance Matter with Small Groups in Massive Classes

Direct answers for search queries in the long tail

Twitch crowdsourcing: crowd contributions in short bursts of time

Street-Level Algorithms: A Theory at the Gaps Between Policy and Decisions