Implementing AI

Kubernetes—the container orchestration engine used by all of the top technology companies—was built from the ground up to run and manage highly distributed workloads on huge clusters. Thus, it provides a solid foundation for model development. Daniel Whitenack demonstrates how to easily deploy and scale AI/ML workflows on any infrastructure using Kubernetes. Read more.

From social network photo filters to self-driving cars, computer vision has brought applied deep learning to the masses. Built by the pioneers of computer vision software, PyTorch enables developers to rapidly build computer vision models. Mo Patel and David Mueller offer an overview of computer vision fundamentals and walk you through using PyTorch to build computer vision applications. Read more.

High-resolution land cover maps help quantify long-term trends like deforestation and urbanization but are prohibitively costly and time intensive to produce. Mary Wahl and Banibrata De demonstrate how to use Microsoft’s Cognitive Toolkit and Azure cloud resources to produce land cover maps from aerial imagery by training a semantic segmentation DNN—both on single VMs and at scale on GPU clusters. Read more.

Xiaoyong Zhu, Gheorghe Iordanescu, Wilson Lee, and Ivan Tarapov walk you through building a deep learning model and intelligent applications on edge devices running iOS, Android, and Windows, using a working example that helps clinicians in areas with less access to radiologists identify possible lung diseases. Read more.

Ray is a new distributed execution framework for reinforcement learning applications. Ion Stoica, Robert Nishihara, and Philipp Moritz lead a deep dive into Ray, walking you through its API and system architecture and sharing application examples, including several state-of-the-art reinforcement learning algorithms. Read more.

Simon Crosby details an architecture for learning on time series data using edge devices, based on the distributed actor model. This approach flies in the face of the traditional wisdom of cloud-based, big-data solutions to ML problems. You'll see that there are more than enough resources at “the edge” to cost-effectively analyze, learn from, and predict from streaming data on the fly. Read more.

Uber has implemented an ML and NLP system that suggests the most likely solutions to a ticket to its customer support representatives, making them faster and more accurate while providing a better user experience. Piero Molino, Huaixiu Zheng, and Yi-Chia Wang describe how Uber built the system with traditional and deep learning models and share the lessons learned along the way. Read more.

Magnus Hyttsten and Priya Gupta demonstrate how to perform distributed TensorFlow training using the Keras high-level APIs. They walk you through TensorFlow's distributed architecture, how to set up a distributed cluster using Kubeflow and Kubernetes, and how to distribute models created in Keras. Read more.

Over the last few years, convolutional neural networks (CNN) have risen in popularity, especially for computer vision. Anirudh Koul explains how to bring the power of convolutional neural networks and deep learning to memory- and power-constrained devices like smartphones. Read more.

KC Tung explains why LSTM provides great flexibility to model the consumer touchpoint sequence problem in a way that allows just-in-time insights about an advertising campaign's effectiveness across all touchpoints (channels), empowering advertisers to evaluate, adjust, or reallocate resources or investments in order to maximize campaign effectiveness. Read more.

Deep learning is often called a black box, so how do you diagnose and fix problems in a deep neural network (DNN)? Avesh Singh and Kevin Wu explain how they systematically debugged DeepHeart, a DNN that detects cardiovascular disease from heart rate data. You'll leave with an arsenal of tools for debugging DNNs, including Jacobian analysis, TensorBoard, and DNN unit tests. Read more.

In spite of the enormous excitement about the potential of deep learning, several key challenges—from prohibitive hardware requirements to immature software offerings—are impeding its widespread enterprise adoption. Evan Sparks and Ameet Talwalkar detail fundamental challenges facing organizations looking to adopt deep learning and present novel solutions to overcome several of them. Read more.

Shaoshan Liu explains how PerceptIn built a reliable autonomous vehicle with a total cost under $10,000. Read more.

Jian Wu discusses an end-to-end engineering project to train and evaluate deep Q-learning models for targeting sequential marketing campaigns using the 10-fold cross-validation method. Jian also explains how to evaluate trained DQN models with neural network-based baseline models and shows that trained deep Q-learning models generally produce better-optimized long-term rewards. Read more.

Complex diseases like Alzheimer’s cannot be cured by pharmaceutical or genetic sciences alone, and current treatments and therapies lead to mixed successes. Ayin Vala explains how to use the power of big data and AI to treat challenging diseases with personalized medicine, which takes into account individual variability in medicine intake, lifestyle, and genetic factors for each patient. Read more.

Earlier this year, Amazon, Facebook, and Microsoft partnered to create the Open Neural Network Exchange (ONNX)—an open format to represent deep learning models. Sarah Bird explains in detail how the ONNX framework can help you take AI from research to reality as quickly as possible. Read more.

David Martinez discusses an AI canonical architecture suitable for a number of different classes of applications and shares examples focused on cybersecurity to illustrate an application area that benefits from an end-to-end AI architecture. Read more.

What does it mean to explain a machine learning model, and why is it important? Armen Donigian addresses those questions while discussing several modern explainability methods, including traditional feature contributions, LIME, and DeepLift. Each of these techniques offers a different perspective, and their clever application can reveal new insights and solve business requirements. Read more.

Captricity has deployed a machine learning pipeline that can read handwriting at human-level accuracy. Ramesh Sridharan discusses the big ideas the company learned building and deploying this system, using data to identify specific problems to solve using AI and to evaluate and validate the algorithm itself and the overall system once deployed. Read more.

Join Ari Kaplan, Erin LeDell, Chris Coad, and David Kearns to see where AI meets business intelligence, as they explore the latest ML technologies and concepts powering today's decisions, including Hortonworks, Aginity Amp, H2O.ai, IBM Data Science Experience, and more—using real-life baseball data to illustrate the concepts. Read more.

Using novel techniques in model-free deep reinforcement learning and control theory, Cathy Wu explores and quantifies the potential impact of a small fraction of automated vehicles on low-level traffic flow dynamics, such as congestion on a variety of important traffic contexts. Read more.

In recent years, there's been a quantum leap in the performance of AI, as deep learning made its mark in areas from speech recognition to machine translation and computer vision. However, as artificial intelligence becomes increasingly popular, data privacy issues also gain traction. Yishay Carmiel reviews these issues and explains how they impact the future of deep learning development. Read more.

With the help of better software, cloud infrastructure, and pretrained networks, AI models have become easier to build. But once your solution veers from a common path, hidden challenges in reproducibility and implementation arise. Brian Dalessandro and Chris Smith share their experience and lessons learned while building a computer vision and OCR app for reading and classifying insurance cards. Read more.

Facebook's strength in AI innovation comes from its ability to quickly bring cutting-edge research into large-scale production using a multifaceted toolset. Joseph Spisak explains how PyTorch 1.0 helps to accelerate the path from research to production by making AI development more seamless and interoperable. Read more.

Personalization is a common theme in social networks and ecommerce businesses. Personalization at Uber involves an understanding of how each driver and rider is expected to behave on the platform. Ankit Jain explains how Uber employs deep learning using LSTMs and its huge database to understand and predict the behavior of each and every user on the platform. Read more.

Building complex, real-world reinforcement learning systems requires leveraging techniques such as curriculum learning, hierarchical RL, and reward shaping. Mark Hammond explores many of these techniques and illustrates how they can be effectively combined into a comprehensive machine teaching program. Read more.

Clara’s human-in-the-loop scheduling service combines the precision of machine intelligence and the judgement of an expert team. Jason Laska explores the trade-offs between text annotations defined for fast data entry and those meant solely for training machine learning models, using the application of DateTime text as it pertains to meeting-attendee availability to guide the discussion. Read more.

Session by Andrew Feldman Read more.

Online fraud is often orchestrated by organized crime rings, who use malicious user accounts to actively target modern online services for financial gain. Ting-Fang Yen shares a real-time, scalable fraud detection solution backed by deep learning and built on Spark and TensorFlow and demonstrates how the system outperforms traditional solutions such as blacklists and machine learning. Read more.

Would you believe that AI inferencing can be done on chips that cost less than a dollar? uTensor, a custom TensorFlow runtime for microcontrollers (MCUs), lets you do just that. Neil Tan offers an overview of uTensor, the first framework to streamline model deployments on MCUs, allowing you to push AI to the edge rather than sending everything to the cloud. Read more.

Abhishek Tayal offers insight into how Twitter's ML platform team, Cortex, is developing models, related tooling, and infrastructure with the objective of making entity embeddings a first-class citizen within Twitter's ML platform. Abhishek also shares success stories on how developing such an ecosystem increases efficiency and productivity and leads to better outcomes across product ML teams. Read more.

Google's AIY Projects kits bring Google's machine learning algorithms to developers with limited experience in the field, allowing them to prototype machine learning applications and smart hardware more easily. Alasdair Allan walks you through setting up and building the kits and demonstrates how to use the kits' Python SDK for machine learning both in the cloud and locally on a Raspberry Pi. Read more.

Lukas Biewald offers an overview of real-world deployments of deep learning models at companies like Home Depot, P&G, Coca-Cola, and Uber, covering practical issues that come up training and iterating on models and problems that can arise post deployment. Lukas also discusses recent research that is directly relevant to industry such as active learning, multitask learning, and transfer learning. Read more.

Labhesh Patel explains how deep learning is informing Jumio's computer vision through smarter data extraction, fraud detection, and risk scoring and how Jumio is leveraging massive datasets and human review to dramatically improve the accuracy of its machine learning algorithms to detect bogus IDs and streamline the verification process of legitimate documents. Read more.

Over 400M people worldwide have some sort of speech or hearing disorder that prevents them from participating in the job market. Goodman Gu offers an overview of Stride4All, an initiative using AI to open work up for disabled people and empower them for teamwork, and showcases a prototype that uses deep learning and computer vision technologies for gesture recognition of American Sign Language. Read more.

Noah Schwartz explores the most recent advances in cooperative learning systems, including distributed and federated learning systems for real-world, edge-based AI. He also considers the pros and cons of multi-agent systems and demonstrates how Quorum AI is working to bridge the gap with the Quorum AI Framework. Read more.

For developers building a bot or virtual agent, the critical question is which bot to build and why? Today, most can’t answer it without a manual intent discovery process, largely based on guesswork, that uncovers only a percentage of possible opportunities. Ofer Ronen demonstrates techniques, based on machine learning, for faster, more efficient intent discovery. Read more.

Predicting driver intention and behavior is of great importance for the planning and decision-making processes of autonomous driving vehicles. Zhou Xing shares a methodology that can be used to build and train a predictive driver system, helping to learn on-road drivers' intentions, behaviors, associated risks, etc. Read more.

Vas Chellappa explains how to keep your GPUs fed with data as you train the next generation of deep learning architectures and shares a new benchmark suite for evaluating and tuning input pipelines. Read more.