Machine Learning for DevOps Summit schedule

The Worst Game of Telephone Ever

Telephone is the party game where a phrase is whispered from the first player to the last until it spoken out loud to reveal how much the message has changed in transit. Data scientists often create models in R, excel, or even with pencil on paper. What happens when these models and algorithms are handed off to a development team to implement? How often is the final product actually doing what was originally intended? Is this the worst game of telephone ever? Can devOps practices help avoid these problems? What can data science & machine learning do for devOps? Please join Chris Corriere, a devOps advocate with SJ Technologies, as he shares stories about failure, success, trust, communication, Nash games, automation, and machine learning.

Map Reduce for Prioritizing a Backlog The Prisoners' Dilemma & The Stag Hunt The Difference between Complicated & Complex Domains Moving from Maps to Models devOps Dojo Practices for Machine Learning

Chris Corriere has been working with data, phones, networks and writing software for over twenty years. His background in mathematics and engineering has allowed him to adapt to new and industry-specific technologies and provided many unique consulting opportunities. As a devOps professional Chris is committed to culture, automation, learning, sharing, and having a good time while getting work done. Chris is currently a senior devOps advocate with SJ Technologies focused on dojo practices, mapping, & complexity science.

Intelligent Decisions for DevOps Processes: AI in the Service of Software Engineering

AI and Software Engineering have always been in tandem throughout the history of Computer Science. AI algorithms have been implemented in exceptionally useful tools and products using well-established software processes. On the other side, AI has frequently contributed smart analytics in supporting software development. Now, we enter a new era for both disciplines, where speed, volume and accuracy are crucial challenges. Now, the synergy between the two disciplines is as pertinent as ever. In this talk, I will explore this collaboration mainly from the perspective of Software Engineering and show examples of AI contributing smart methods and analytics to support modern software processes.

Marios Fokaefs is an Assistant Professor at École Polytechnique de Montréal, Canada. He holds a BSc in Applied Informatics from the University of Macedonia, Greece, a MSc and a PhD in Software Engineering from the Department of Computing Science, University of Alberta, Canada. His research interests include DevOps, software evolution, change management, service-oriented architectures, cloud systems, Internet-of-Things and software engineering economics. Dr Fokaefs is an IEEE Member and an IBM CAS Faculty. His research has been supported and funded by IBM, AITF from Alberta, ORF, OCE and SOSCIP from Ontario, as well as the Natural Sciences and Engineering Research Council of Canada.

Wasted Gold Mine & What Data Can Do To DevOps

As CTO of CloudBees and the creator of Jenkins, I get to see lots of real-world software development. Our automation in software development is sufficiently broad that it is producing lots of data, but by and large most of those are simply thrown away. Yet at the same time, the management is feeling like they are flying blind because they have little insight! In this talk, Kohsuke will discuss how we collectively seem to miss the golden opportunity to improve the software development process itself, based on data. IOW, learning is lacking at the organizational level, let alone "machine" learning!

Kohsuke Kawaguchi is the creator of Jenkins. He is a well-respected developer and popular speaker at industry and Jenkins community events. Kawaguchi’s sensibilities in creating Jenkins and his deep understanding of how to translate its capabilities into usable software have also had a major impact on CloudBees’ strategy as a company. Before joining CloudBees, Kawaguchi was with Sun Microsystems and Oracle, where he worked on a variety of projects and initiated the open source work that led to Jenkins.

Demystifying Neural Networks

Neural Networks are at the cutting edge of Machine Learning, and Artificial Intelligence. Often the black-box reference made to explain NN's leads to folks with little experience in Machine Learning, statistics or big data to become intimidated. There is a powerful synergy between DevOps and Neural Networks, NN's that use automatic feature extraction give exponential leverage in the processing of monitoring production data, identifying patterns or anomalies by analyzing application log events, and verifying production deployments.

In this talk I will walk you through how we can "Demystify Neural Nets". By utilizing open-source libraries like TensorFlow, and Keras, I will demonstrate how one can build a neural network in no time. The specific focus for this presentation will be Long Short-term Memory neural networks (LSTM), that are extremely valuable in building models for time-series data like system metrics data, and application log data. This presentation is geared towards folks interested in the field of ML, both beginners and intermediate levels.

Boshika has four plus years of experience working as a full-stack engineer in San Francisco Bay Area and Los Angeles. She currently works for Commercial Tech at Capital One, where she is building microservices in Golang for commercial document migration. She is also part of the agenda at Commercial Tech to apply machine learning to solve business cases related to data classification, and data extraction. Boshika became fascinated by the field of data science and machine learning while doing bench research at Stanford University, where she was using ML algorithms to analyze large scale genome sequencing data. She is also currently pursuing her Master’s in Data Science at Johns Hopkins University.

Forecasting Risk using CICD and Machine Learning

The software development process does not have a singular formula for success. There are many factors that influence this process such as code management, quality analysis, testing strategies to name a few. Pinpointing the features that correlate to the success of a deployment are subjective to a developer’s personal style of software engineering and testing. I will be presenting a project that implemented a machine learning component to analyze build features and gauge them against a pool of build records from a project and organization perspective.

Omari S. Felix, a DevOps Engineer at Capital One where he implements DevOps tools and guide teams on utilizing techniques designed to improve their software development and delivery. A graduate from Virginia State University and attended N.C. A&T State University. He has experience in multiple technology spaces include mobile development, automation, and machine learning to name a few.

Machine Learning Meets DevOps

Today’s mandate for faster business innovation, faster response to changes in the market, and faster development of new products demand a new paradigm for software development. DevOps is a set of practices that aim to decrease the time between changing a system in Development, and transferring the change to the Operation environment, and exploiting the Operation data back in the Development. DevOps practices are typically relying on large amount of data coming from Operation. The amount of data depends on the architectural style, the underlying development technologies and deployment infrastructure. However, in order to make effective decisions in Development, e.g., architecture changes in continuous delivery pipelines, there has to be efficient processing of big data in Operation. In this situation where data streams are increasingly large-scale, dynamical and heterogeneous, mathematical and algorithmic creativity are required to bring statistical methodology to bear. Statistical machine learning can fill the gap between operation and development with more efficient analytical techniques. Such analytical techniques can provide more deep knowledge and can uncover the underlying patterns in the operational data, e.g., to detect anomalies in the operation or detect performance anti-patterns. This knowledge can be very practical if detected ontime to refactor the development artifacts including code, architecture and deployment.

In this talk, I will start motivating the necessity of data-driven analytics for generating feedback to Dev from Ops based on my previous experience with large-scale big data systems in industry. I will present our recent work on configuration tuning of big data software, where we primarily applied Bayesian Optimization and Gaussian Processes to effectively find optimum configurations. I will also talk about transfer learning to exploit complimentary and cheap information (e.g., past measurements in a continuous delivery pipeline regarding early versions of the system) to enable learning accurate models efficiently and with considerably less cost. Results show that despite the high cost of measurement on the real system, learning performance models can become surprisingly cheap as long as certain properties are reused across environments. In the second half of the talk, I will present empirical evidence, which lays a foundation for a theory explaining why and when transfer learning works by showing the similarities of performance behavior across environments. I will present observations of environmental changes‘ impacts (such as changes to hardware, workload, and software versions which are predominant in DevOps) for a selected set of configurable systems from different domains to identify the key elements that can be exploited for transfer learning. These observations demonstrate a promising path for building efficient, reliable, and dependable software systems.

Pooyan Jamshidi is an Assistant Professor at the University of South Carolina. Prior to his current position, he was a research associate at Carnegie Mellon University (2016-2018) and Imperial College London (2014-2016), where he primarility worked on transfer learning for performance analyses of highly-configurable systems including robotics and big data systems. He holds a Ph.D. from Dublin City University (2010-2014). Pooyan's general research interests are at the intersection of software engineering, systems, and machine learning, and his focus is primarily in the areas of distributed machine learning. Pooyan has spent 7 years in software industry before his PhD.

Deploying Scalable ML Models in the Enterprise

After massive investments in collecting and cleaning data, then training Machine Learning models—Enterprises discover the big challenges in deploying models to production and managing their growing portfolio of ML models. This talk will cover the strategic and technical hurdles each company must overcome and the best practices we've developed while deploying over 5,000 ML models for 75,000 engineers.

Diego Oppenheimer, founder and CEO of Algorithmia, is an entrepreneur and product developer with an extensive background in all things data. Prior to founding Algorithmia he designed, managed and shipped some of Microsoft’s most used data analysis products including Excel, Power Pivot, SQL Server and Power BI. Diego holds a Bachelors degree in Information Systems and a Masters degree in Business Intelligence and Data Analytics from Carnegie Mellon University.

Improving Adobe Experience Cloud Services Dependability with Machine Learning

Adobe Experience Cloud is a collection of best-in-class solutions for marketing, analytics, advertising, and commerce. All integrated on a cloud platform for a single experience system of record. The Adobe Experience Cloud's SRE team works hand-in-hand with the Product and Engineering teams to build dependable services. In this presentation you will learn how the team leverage Adobe's artificial intelligence and machine learning engine, to first, build predictive auto-scaling and self-healing services. Second, to provide insight and automate risk classification of production changes to reduce impact on services availability.

This talk will discuss work and finding by Adobe's SRE using LSTM, ETS, Linear Regression, Boosted Tree, Multi-Layer Perceptron, and Support Vector Models.

Nicolas Brousse, a Cloud Technology Leader, became Director of Operations Engineering at Adobe (NASDAQ: ADBE) after the acquisition of TubeMogul (NASDAQ: TUBE). As TubeMogul's sixth employee and first operations hire, Nicolas has built and grown Adobe/TubeMogul's infrastructure over the past ten years from several machines to over eight thousand servers that handle ±350 billions requests per day for clients like Allstate, Chrysler, Heineken and Hotels.com. Adept at adapting quickly to ongoing business needs and constraints, Nicolas leads a global team of site reliability engineers, cloud engineers, software engineers, security engineers, and database architects that build, manage, and monitor Adobe Advertising Cloud's infrastructure 24/7 and adhere to "DevOps" methodology. https://nicolas.brousse.info/

Large Scale Inference Generation Using Jenkins/Blue Ocean and GPU Cluster

For our ecommerce use cases, we have several catalogs in O(10⁶) items scale for which we need to generate embeddings on a daily basis. The catalogs are constantly expanding and updated. These embeddings could be created from TensorFlow models trained with catalog images and/or meta data for the items. For different use cases, the TensorFlow models and/or the catalogs are different, and we need a common platform for large scale inference generation. We are a very small team of 2 data scientists and 1 engineer. We have access to GPU cluster with different types of Nvidia GPUs, and these GPU servers are used primarily for training purpose.

Among the challenges we face are,

Selecting idle GPU servers with matching docker containers for them Distributing the workload of inference generation to idle GPU servers Automating the integration tests for model server and API server (model version matching the embeddings, as well as supported by the docker containers in production) Working with Walmart infrastructure for docker containers (security constraints, no nvidia-docker on GPU servers etc.)

Our solution consists of a Jenkins cluster with all the GPU servers added as worker nodes. We use parallel Blue Ocean pipeline to distribute the workload. For various Nvidia GPUs, we maintain docker containers that support the corresponding CUDA and Nvidia driver versions. We take this approach because it is easy to add new GPU servers, and also upgrade the docker images for GPU servers. We use object storage for both models and embeddings, which are stored in the same subfolders, in order to ensure that the generated embeddings and the TensorFlow model are ready for deployment together.

In my talk, I would also touch on the DevOps aspects of building docker images do’s and don’ts (keeping the docker images lean, learnings from building TensorFlow model server subject to enterprise constraints etc.) and deployment to Kubernetes (using init-containers, rolling updates etc.)

Future directions we are exploring:

TensorRT inference server Kubernetes on Nvidia GPUs Google TPU for inference

Binwei Yang is an engineer, a hacker, and a hustler. He is passionate about becoming a lifelong learner and equipping youth in underserved communities with high-tech skills. Binwei graduated from University of Southern California with a Master in Computer Engineering and Ph.D. in Physics, and has more than 20 years of professional experience of creating massively scalable customer-facing applications. He currently works on computer vision as a principal engineer for Walmart Labs.

AIOps Solution Enables Self-Healing Hybrid IT

Enterprises running their digital business services in hybrid heterogeneous IT environment require visibility and proactive insights into problem areas. Operational data volume and number of data sources have increased tremendously in modern data centers further worsening the problem of visibility and correlation. Artificial Intelligence for IT Operations (AIOps) leverages Big Data, Machine Learning and AI techniques to deliver proactive and predictive insights, recommendations and remediation. By combining the power of auto-discovery and correlation with ML and AI, AIOps solution suppresses events to reduce noise for faster root cause analysis, predicts business incidents, detects multivariate anomaly scenarios and enables enterprises with top-down visibility from business KPIs to application to infrastructure. This talk will discuss how AIOps solution modernizes and unifies IT Operations using its auto-discovery, correlation, machine learning and AI techniques.

Bishnu is the CTO of FixStream, an emerging AIOps startup in Silicon Valley where he guides FixStream's overall technology, product strategy and roadmap, R&D and engagement with technology partners, enterprise customers and service providers. Prior to FixStream, Bishnu spent thirteen years at AT&T in various executive architect and strategist positions where he led technology programs in the domains of Cloud, Enterprise Architecture, Big Data, Mobile, DevOps etc. Bishnu has a flair for all things Cloud technology, Big Data and, ML and APIs. Bishnu is very passionate about cutting-edge technologies and is very active in various industry recognized communities, blogs and forums. Bishnu is an active member of the Forbes Technology Council (ForbesTechCouncil.com) where he has been actively participating and contributing since 2016.

The Worst Game of Telephone Ever

Telephone is the party game where a phrase is whispered from the first player to the last until it spoken out loud to reveal how much the message has changed in transit. Data scientists often create models in R, excel, or even with pencil on paper. What happens when these models and algorithms are handed off to a development team to implement? How often is the final product actually doing what was originally intended? Is this the worst game of telephone ever? Can devOps practices help avoid these problems? What can data science & machine learning do for devOps? Please join Chris Corriere, a devOps advocate with SJ Technologies, as he shares stories about failure, success, trust, communication, Nash games, automation, and machine learning.

Map Reduce for Prioritizing a Backlog The Prisoners' Dilemma & The Stag Hunt The Difference between Complicated & Complex Domains Moving from Maps to Models devOps Dojo Practices for Machine Learning

Chris Corriere has been working with data, phones, networks and writing software for over twenty years. His background in mathematics and engineering has allowed him to adapt to new and industry-specific technologies and provided many unique consulting opportunities. As a devOps professional Chris is committed to culture, automation, learning, sharing, and having a good time while getting work done. Chris is currently a senior devOps advocate with SJ Technologies focused on dojo practices, mapping, & complexity science.

Anurag is currently working as a Cloud Software Engineer in the Cloud Infrastructure Team at the Geophysics Technology Centre, Schlumberger. His work entails developing and maintaining scalable backend services and CI/CD pipelines. Before this, Anurag earned his Masters in Computer Science from the University of Florida and his Bachelors in Computer Engineering from the University of Pune. His other research interests include Information Security, Machine Learning, and IoT. While he’s not developing code, Anurag likes to travel and explore, he’s an avid photographer and tries to combine his hobbies as a Google Maps Local Guide.

Dynamic Banking - A five-year roadmap to excellence in AI

Andrew Ng famously stated that “AI is the new electricity”, whilst the jury is still out on this one - it is important to note there has already been at least two seperate AI renaissances in the past that failed to reach escape velocity*. However it is without a doubt that the next AI revolution has begun, but this time far greater in scope and scale than ever before. That said learnings have been made, and it has become apparent that a robust framework is essential for successful adoption across the various industries - this is where AI meets DevOps.

In this presentation I showcase a one size fits all, five-year roadmap to guide businesses big and small into a successful AI transformation. I also explore Dynamic Banking - the first major wave of AI finance centric innovations from slim predictive fraud engines to frictionless banking IPB micro-apps. Finally I report on the progress across the globe towards the second major wave of AI, and what this means for finance as a whole.

• Significant investment from both the US, UK and the EU was funnelled into AI research and development in the 1960s and 1980s, in both cases funding and interest was abruptly withdrawn due to limitations of the technology. Adam McMurchie is leader in Devops and an A.I expert working in the banks SAO platform on the forefront of technology development in finance. With a broad exposure to a range of technologies, Adam drives an ethos of simplification, cloud agnosticism and specialises in spotting the next trends in fin tech. Additionally, Adam also has a background in science with a physics degree specialising in NeuroComputing and is a polyglot linguist & seasoned translator. Adam has pooled these skills to deliver full stack novel solutions from tensor flow driven mobile apps, to personalized banking chatbots. Adam also develops apps designed around the ethos of Social Utility, including Flood/Storm reporting, EV Vehicle bay monitoring and preservation of endangered languages.

David Pierce is a Senior Software Engineer passionate about bridging Analytics & Engineering by developing Comprehensive Data Strategies on Heterogeneous Infrastructure. By building Modular Data Pipelines and treating infrastructure as a software problem, Data SMEs can leverage Automated Machine Learning and Self-Healing Data to continually evolve their applications and solutions through self-service tools.

Giulia Toti is an Assistant Instructional faculty at the University of Houston. She offers several core classes in the CS curriculum and a selection of courses designed in collaboration with the Hewlett Packard Enterprise Data Science Institute to increase the presence of machine learning in the curriculum.

Giulia obtained a PhD in computer science from the University of Houston in 2016. Upon graduation, Giulia joined the Addictions Department of the King’s College London as postdoctoral researcher. There, she worked on mining of large electronic health records databases and on the development of risk prediction models. In Fall 2017 Giulia came back to the University of Houston as instructional faculty. Besides teaching, she works to promote research among undergraduates and is one of the mentors for the Summer Research Experience for Undergraduates (REU) program. She is also a member of The National Center for Women in Information Technology's (NCWIT) and involved in increasing women representation in the department.

Deploying Scalable ML Models in the Enterprise

After massive investments in collecting and cleaning data, then training Machine Learning models—Enterprises discover the big challenges in deploying models to production and managing their growing portfolio of ML models. This talk will cover the strategic and technical hurdles each company must overcome and the best practices we've developed while deploying over 5,000 ML models for 75,000 engineers.

Diego Oppenheimer, founder and CEO of Algorithmia, is an entrepreneur and product developer with an extensive background in all things data. Prior to founding Algorithmia he designed, managed and shipped some of Microsoft’s most used data analysis products including Excel, Power Pivot, SQL Server and Power BI. Diego holds a Bachelors degree in Information Systems and a Masters degree in Business Intelligence and Data Analytics from Carnegie Mellon University.