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

When you share content in an online social network, who is listening? Users have scarce information about who actually sees their content, making their audience seem invisible and difficult to estimate. However, understanding this invisible audience can impact both science and design, since perceived audiences influence content production and self-presentation online. In this paper, we combine survey and large-scale log data to examine how well users' perceptions of their audience match their actual audience on Facebook. We find that social media users consistently underestimate their audience size for their posts, guessing that their audience is just 27% of its true size. Qualitative coding of survey responses reveals folk theories that attempt to reverse-engineer audience size using feedback and friend count, though none of these approaches are particularly accurate. We analyze audience logs for 222,000 Facebook users' posts over the course of one month and find that publicly visible signals --- friend count, likes, and comments --- vary widely and do not strongly indicate the audience of a single post. Despite the variation, users typically reach 61% of their friends each month. Together, our results begin to reveal the invisible undercurrents of audience attention and behavior in online social networks.

/pdf/quantifying-the-invisible-audience-in-social-networks-5g26h4mgf9.pdf

Quantifying the invisible audience in social networks

Interactive systems must respond to user input within seconds. Therefore, to create realtime crowd-powered interfaces, we need to dramatically lower crowd latency. In this paper, we introduce the use of synchronous crowds for on-demand, realtime crowdsourcing. With synchronous crowds, systems can dynamically adapt tasks by leveraging the fact that workers are present at the same time. We develop techniques that recruit synchronous crowds in two seconds and use them to execute complex search tasks in ten seconds. The first technique, the retainer model, pays workers a small wage to wait and respond quickly when asked. We offer empirically derived guidelines for a retainer system that is low-cost and produces on-demand crowds in two seconds. Our second technique, rapid refinement, observes early signs of agreement in synchronous crowds and dynamically narrows the search space to focus on promising directions. This approach produces results that, on average, are of more reliable quality and arrive faster than the fastest crowd member working alone. To explore benefits and limitations of these techniques for interaction, we present three applications: Adrenaline, a crowd-powered camera where workers quickly filter a short video down to the best single moment for a photo; and Puppeteer and A|B, which examine creative generation tasks, communication with workers, and low-latency voting.

Crowds in two seconds: enabling realtime crowd-powered interfaces

Prototyping is the pivotal activity that structures innovation, collaboration, and creativity in design. Prototypes embody design hypotheses and enable designers to test them. Framin design as a thinking-by-doing activity foregrounds iteration as a central concern. This paper presents d.tools, a toolkit that embodies an iterative-design-centered approach to prototyping information appliances. This work offers contributions in three areas. First, d.tools introduces a statechart-based visual design tool that provides a low threshold for early-stage prototyping, extensible through code for higher-fidelity prototypes. Second, our research introduces three important types of hardware extensibility - at the hardware-to-PC interface, the intra-hardware communication level, and the circuit level. Third, d.tools integrates design, test, and analysis of information appliances. We have evaluated d.tools through three studies: a laboratory study with thirteen participants; rebuilding prototypes of existing and emerging devices; and by observing seven student teams who built prototypes with d.tools.

/pdf/reflective-physical-prototyping-through-integrated-design-11ekdiahfp.pdf

Reflective physical prototyping through integrated design, test, and analysis

We present two studies of online ephemerality and anonymity based on the popular discussion board /b/ at 4chan.org: a website with over 7 million users that plays an influential role in Internet culture. Although researchers and practitioners often assume that user identity and data permanence are central tools in the design of online communities, we explore how /b/ succeeds despite being almost entirely anonymous and extremely ephemeral. We begin by describing /b/ and performing a content analysis that suggests the community is dominated by playful exchanges of images and links. Our first study uses a large dataset of more than five million posts to quantify ephemerality in /b/. We find that most threads spend just five seconds on the first page and less than five minutes on the site before expiring. Our second study is an analysis of identity signals on 4chan, finding that over 90% of posts are made by fully anonymous users, with other identity signals adopted and discarded at will. We describe alternative mechanisms that /b/ participants use to establish status and frame their interactions.

/pdf/4chan-and-b-an-analysis-of-anonymity-and-ephemerality-in-a-4ze25feudd.pdf

4chan and /b/: An Analysis of Anonymity and Ephemerality in a Large Online Community

By lowering the costs of communication, the web promises to enable distributed collectives to act around shared issues. However, many collective action efforts never succeed: while the web's affordances make it easy to gather, these same decentralizing characteristics impede any focus towards action. In this paper, we study challenges to collective action efforts through the lens of online labor by engaging with Amazon Mechanical Turk workers. Through a year of ethnographic fieldwork, we sought to understand online workers' unique barriers to collective action. We then created Dynamo, a platform to support the Mechanical Turk community in forming publics around issues and then mobilizing. We found that collective action publics tread a precariously narrow path between the twin perils of stalling and friction, balancing with each step between losing momentum and flaring into acrimony. However, specially structured labor to maintain efforts' forward motion can help such publics take action.

/pdf/we-are-dynamo-overcoming-stalling-and-friction-in-collective-5f5tqqsq5s.pdf

Michael S. Bernstein

Papers

Quantifying the invisible audience in social networks

Crowds in two seconds: enabling realtime crowd-powered interfaces

Reflective physical prototyping through integrated design, test, and analysis

4chan and /b/: An Analysis of Anonymity and Ephemerality in a Large Online Community

We Are Dynamo: Overcoming Stalling and Friction in Collective Action for Crowd Workers