O'Reilly Data Show Podcast
O'Reilly Data Show Podcast
O'Reilly Media
The O'Reilly Data Show Podcast explores the opportunities and techniques driving big data, data science, and AI.
Machine learning for operational analytics and business intelligence
In this episode of the Data Show, I speak with Peter Bailis, founder and CEO of Sisu, a startup that is using machine learning to improve operational analytics. Bailis is also an assistant professor of computer science at Stanford University, where he conducts research into data-intensive systems and where he is co-founder of the DAWN Lab. We had a great conversation spanning many topics, including: His personal blog, which contains some of the best explainers on emerging topics in data management and distributed systems. The role of machine learning in operational analytics and business intelligence. Machine learning benchmarks—specifically two recent ML initiatives that he’s been involved with: DAWNBench and MLPerf. Trends in data management and in tools for machine learning development, governance, and operations. Related resources: “Setting benchmarks in machine learning”: Dave Patterson, Peter Bailis, and other industry leaders discuss how MLPerf will define an entire suite of benchmarks to measure performance of software, hardware, and cloud systems. “The quest for high-quality data” “RISELab’s AutoPandas hints at automation tech that will change the nature of software development” Jeff Jonas on “Real-time entity resolution made accessible” “What are model governance and model operations?” “We need to build machine learning tools to augment machine learning engineers”
Oct 10, 2019
51 min
Machine learning and analytics for time series data
In this episode of the Data Show, I speak with Arun Kejariwal of Facebook and Ira Cohen of Anodot (full disclosure: I’m an advisor to Anodot). This conversation stemmed from a recent online panel discussion we did, where we discussed time series data, and, specifically, anomaly detection and forecasting. Both Kejariwal (at Machine Zone, Twitter, and Facebook) and Cohen (at HP and Anodot) have extensive experience building analytic and machine learning solutions at large scale, and both have worked extensively with time-series data. The growing interest in AI and machine learning has not been confined to computer vision, speech technologies, or text. In the enterprise, there is strong interest in using similar automation tools for temporal data and time series. We had a great conversation spanning many topics, including: Why businesses should care about anomaly detection and forecasting; specifically, we delve into examples outside of IT Operations & Monitoring. (Specialized) techniques and tools for automating some of the relevant tasks, including signal processing, statistical methods, and machine learning. What are some of the key features of an anomaly detection or forecasting system. What lies ahead for large-scale systems for time series analysis. Related resources: “Product management in the machine learning era” – a new tutorial at the Artificial Intelligence Conference in London “One simple chart: Who is interested in Apache Pulsar?” Ira Cohen: “Semi-supervised, unsupervised, and adaptive algorithms for large-scale time series” “Got speech? These guidelines will help you get started building voice applications” “RISELab’s AutoPandas hints at automation tech that will change the nature of software development” Ameet Talwalker: “How to train and deploy deep learning at scale”
Sep 26, 2019
40 min
Understanding deep neural networks
In this episode of the Data Show, I speak with Michael Mahoney, a member of RISELab, the International Computer Science Institute, and the Department of Statistics at UC Berkeley. A physicist by training, Mahoney has been at the forefront of many important problems in large-scale data analysis. On the theoretical side, his works spans algorithmic and statistical methods for matrices, graphs, regression, optimization, and related problems. On the applications side, he has contributed to systems used for internet and social media analysis, social network analysis, as well as for a host of applications in the physical and life sciences. Most recently, he has been working on deep neural networks, specifically developing theoretical methods and practical diagnostic tools that should be helpful to practitioners who use deep learning. Analyzing deep neural networks with WeightWatcher. Image by Michael Mahoney and Charles Martin, used with permission. We had a great conversation spanning many topics, including: The class of problems in big data, machine learning, and data analysis that he has worked on at Yahoo, Stanford, and Berkeley. The new UC Berkeley FODA (Foundations of Data Analysis) Institute. HAWQ (Hessian AWare Quantization of Neural Networks with Mixed-Precision), a new framework for addressing problems pertaining to model size and inference speed/power in deep learning. WeightWatcher: a new open source project for predicting the accuracy of deep neural networks. WeightWatcher stems from a recent series of papers with Charles Martin, of Calculation Consulting. Related resources: “Deep learning at scale: Tools and solutions” – a new tutorial at the Artificial Intelligence Conference in San Jose Ameet Talwalker on “How to train and deploy deep learning at scale” Greg Diamos on “How big compute is powering the deep learning rocket ship” “RISELab’s AutoPandas hints at automation tech that will change the nature of software development” Reza Zadeh on “Scaling machine learning” “Becoming a machine learning company means investing in foundational technologies” “Managing risk in machine learning” “What are model governance and model operations?” “Product management in the machine learning era”: a tutorial at the Artificial Intelligence Conference in San Jose
Sep 12, 2019
39 min
Becoming a machine learning practitioner
In this episode of the Data Show, I speak with Kesha Williams, technical instructor at A Cloud Guru, a training company focused on cloud computing. As a full stack web developer, Williams became intrigued by machine learning and started teaching herself the ML tools on Amazon Web Services. Fast forward to today, Williams has built some well-regarded Alexa skills, mastered ML services on AWS, and has now firmly added machine learning to her developer toolkit. Anatomy of an Alexa skill. Image by Kesha Williams, used with permission. We had a great conversation spanning many topics, including: How she got started and made the transition into a full-fledged machine learning practitioner. We discussed the evolution of ML tools and learning resources, and how accessible they’ve become for developers. How to build and monetize Alexa skills. Along the way, we took a deep dive and discussed some of the more interesting Alexa skills she has built, as well as one that she really admires. Related resources: “Product management in the machine learning era”: a new tutorial session at the Artificial Intelligence Conference in London Cassie Kozyrkov: “Make data science more useful” Kartik Hosanagar: “Algorithms are shaping our lives—here’s how we wrest back control” Francesca Lazzeri and Jaya Mathew: “Lessons learned while helping enterprises adopt machine learning” Jerry Overton: “Teaching and implementing data science and AI in the enterprise” “Becoming a machine learning company means investing in foundational technologies” “Managing risk in machine learning” “What are model governance and model operations?”
Aug 29, 2019
33 min
Labeling, transforming, and structuring training data sets for machine learning
In this episode of the Data Show, I speak with Alex Ratner, project lead for Stanford’s Snorkel open source project; Ratner also recently garnered a faculty position at the University of Washington and is currently working on a company supporting and extending the Snorkel project. Snorkel is a framework for building and managing training data. Based on our survey from earlier this year, labeled data remains a key bottleneck for organizations building machine learning applications and services. Ratner was a guest on the podcast a little over two years ago when Snorkel was a relatively new project. Since then, Snorkel has added more features, expanded into computer vision use cases, and now boasts many users, including Google, Intel, IBM, and other organizations. Along with his thesis advisor professor Chris Ré of Stanford, Ratner and his collaborators have long championed the importance of building tools aimed squarely at helping teams build and manage training data. With today’s release of Snorkel version 0.9, we are a step closer to having a framework that enables the programmatic creation of training data sets. Snorkel pipeline for data labeling. Source: Alex Ratner, used with permission. We had a great conversation spanning many topics, including: Why he and his collaborators decided to focus on “data programming” and tools for building and managing training data. A tour through Snorkel, including its target users and key components. What’s in the newly released version (v 0.9) of Snorkel. The number of Snorkel’s users has grown quite a bit since we last spoke, so we went through some of the common use cases for the project. Data lineage, AutoML, and end-to-end automation of machine learning pipelines. Holoclean and other projects focused on data quality and data programming. The need for tools that can ease the transition from raw data to derived data (e.g., entities), insights, and even knowledge. Related resources: “Product management in the machine learning era”: A tutorial at the Artificial Intelligence Conference in San Jose, September 9-12, 2019. Chris Ré: “Software 2.0 and Snorkel” Alex Ratner: “Creating large training data sets quickly” Ihab Ilyas and Ben Lorica on “The quest for high-quality data” Roger Chen: “Acquiring and sharing high-quality data” Jeff Jonas on “Real-time entity resolution made accessible” “Data collection and data markets in the age of privacy and machine learning”
Aug 15, 2019
40 min
Make data science more useful
In this episode of the Data Show, I speak with Cassie Kozyrkov, technical director and chief decision scientist at Google Cloud. She describes “decision intelligence” as an interdisciplinary field concerned with all aspects of decision-making, and which combines data science with the behavioral sciences. Most recently she has been focused on developing best practices that can help practitioners make safe, effective use of AI and data. Kozyrkov uses her platform to help data scientists develop skills that will enable them to connect data and AI with their organizations’ core businesses. We had a great conversation spanning many topics, including: How data science can be more useful The importance of the human side of data The leadership talent shortage in data science Is data science a bubble? Related resources: “Managing machine learning in the enterprise: Lessons from banking and health care” “Managing risk in machine learning” “What are model governance and model operations?” “Becoming a machine learning company means investing in foundational technologies” Forough Poursabzi Sangdeh: “It’s time for data scientists to collaborate with researchers in other disciplines” Jacob Ward: “How social science research can inform the design of AI systems” “AI and machine learning will require retraining your entire organization” Ihab Ilyas and Ben lorica on “The quest for high-quality data” “Product management in the machine learning era”—a tutorial at the Artificial Intelligence Conference in San Jose
Aug 1, 2019
35 min
Acquiring and sharing high-quality data
In this episode of the Data Show, I spoke with Roger Chen, co-founder and CEO of Computable Labs, a startup focused on building tools for the creation of data networks and data exchanges. Chen has also served as co-chair of O’Reilly’s Artificial Intelligence Conference since its inception in 2016. This conversation took place the day after Chen and his collaborators released an interesting new white paper, Fair value and decentralized governance of data. Current-generation AI and machine learning technologies rely on large amounts of data, and to the extent they can use their large user bases to create “data silos,” large companies in large countries (like the U.S. and China) enjoy a competitive advantage. With that said, we are awash in articles about the dangers posed by these data silos. Privacy and security, disinformation, bias, and a lack of transparency and control are just some of the issues that have plagued the perceived owners of “data monopolies.” In recent years, researchers and practitioners have begun building tools focused on helping organizations acquire, build, and share high-quality data. Chen and his collaborators are doing some of the most interesting work in this space, and I recommend their new white paper and accompanying open source projects. Sequence of basic market transactions in the Computable Labs protocol. Source: Roger Chen, used with permission. We had a great conversation spanning many topics, including: Why he chose to focus on data governance and data markets. The unique and fundamental challenges in accurately pricing data. The importance of data lineage and provenance, and the approach they took in their proposed protocol. What cooperative governance is and why it’s necessary. How their protocol discourages an unscrupulous user from just scraping all data available in a data market. Related resources: Roger Chen: “Data liquidity in the age of inference” Ihab Ilyas and Ben lorica on “The quest for high-quality data” Chris Ré: “Software 2.0 and Snorkel” Alex Ratner on “Creating large training data sets quickly” Jeff Jonas on “Real-time entity resolution made accessible” “Data collection and data markets in the age of privacy and machine learning” Guillaume Chaslot on “The importance of transparency and user control in machine learning”
Jul 18, 2019
39 min
Tools for machine learning development
In this week’s episode of the Data Show, we’re featuring an interview Data Show host Ben Lorica participated in for the Software Engineering Daily Podcast, where he was interviewed by Jeff Meyerson. Their conversation mainly centered around data engineering, data architecture and infrastructure, and machine learning (ML). Here are a few highlights: Tools for productive collaboration A data catalog, at a high level, basically answers questions around the data that’s available and who is using it so an enterprise can understand access patterns. … The term “data catalog” is generally used when you’ve gotten to the point where you have a team of data scientists and you need a place where they can use libraries in a setting where they can collaborate, and where they can share not only models but maybe even data pipelines and features. The more advanced data science platforms will have automation tools built in. … The ideal scenario is the data science platform is not just for prototyping, but also for pushing things to production. Tools for ML development We have tools for software development, and now we’re beginning to hear about tools for machine learning development—there’s a company here at Strata called Comet.ml, and there’s another startup called Verta.ai. But what has really caught my attention is an open source project from Databricks called MLflow. When it first came out, I thought, ‘Oh, yeah, so we don’t have anything like this. Might have a decent chance of success.’ But I didn’t pay close attention until recently; fast forward to today, there are 80 contributors for 40 companies and 200+ companies using it. What’s good about MLflow is that it has three components and you’re free to pick and choose—you can use one, two, or three. Based on their surveys, the most popular component is the one for tracking and managing machine learning experiments. It’s designed to be useful for individual data scientists, but it’s also designed to be used by teams of data scientists, so they have documented use-cases of MLflow where you have a company managing thousands of models and productions.
Jul 3, 2019
39 min
Enabling end-to-end machine learning pipelines in real-world applications
In this episode of the Data Show, I spoke with Nick Pentreath, principal engineer at IBM. Pentreath was an early and avid user of Apache Spark, and he subsequently became a Spark committer and PMC member. Most recently his focus has been on machine learning, particularly deep learning, and he is part of a group within IBM focused on building open source tools that enable end-to-end machine learning pipelines. We had a great conversation spanning many topics, including: AI Fairness 360 (AIF360), a set of fairness metrics for data sets and machine learning models Adversarial Robustness Toolbox (ART), a Python library for adversarial attacks and defenses. Model Asset eXchange (MAX), a curated and standardized collection of free and open source deep learning models. Tools for model development, governance, and operations, including MLflow, Seldon Core, and Fabric for deep learning Reinforcement learning in the enterprise, and the emergence of relevant open source tools like Ray. Related resources: “Modern Deep Learning: Tools and Techniques”—a new tutorial at the Artificial Intelligence conference in San Jose Harish Doddi on “Simplifying machine learning lifecycle management” Sharad Goel and Sam Corbett-Davies on “Why it’s hard to design fair machine learning models” “Managing risk in machine learning”: considerations for a world where ML models are becoming mission critical “The evolution and expanding utility of Ray” “Local Interpretable Model-Agnostic Explanations (LIME): An Introduction” Forough Poursabzi Sangdeh on why “It’s time for data scientists to collaborate with researchers in other disciplines”
Jun 20, 2019
42 min
Bringing scalable real-time analytics to the enterprise
In this episode of the Data Show, I spoke with Dhruba Borthakur (co-founder and CTO) and Shruti Bhat (SVP of Product) of Rockset, a startup focused on building solutions for interactive data science and live applications. Borthakur was the founding engineer of HDFS and creator of RocksDB, while Bhat is an experienced product and marketing executive focused on enterprise software and data products. Their new startup is focused on a few trends I’ve recently been thinking about, including the re-emergence of real-time analytics, and the hunger for simpler data architectures and tools.  Borthakur exemplifies the need for companies to continually evaluate new technologies: while he was the founding engineer for HDFS, these days he mostly works with object stores like S3. We had a great conversation spanning many topics, including: RocksDB, an open source, embeddable key-value store originated by Facebook, and which is used in several other open source projects. Time-series databases. The importance of having solutions for real-time analytics, particularly now with the renewed interest in IoT applications and rollout of 5G technologies. Use cases for Rockset’s technologies—and more generally, applications of real-time analytics. The Aggregator Leaf Tailer architecture as an alternative to the Lambda architecture. Building data infrastructure in the cloud. The Aggregator Leaf Tailer (“CQRS for the data world”): A data architecture favored by web-scale companies. Source: Dhruba Borthakur, used with permission. Related resources: Serverless Streaming Architectures & Algorithms for the Enterprise – a new tutorial on September 24th at Strata Data NYC. “Becoming a machine learning company means investing in foundational technologies” Haoyuan Li: “In the age of AI, fundamental value resides in data” Harish Doddi: “Simplifying machine learning lifecycle management” Eric Jonas: “A Berkeley view on serverless computing” “Specialized tools for machine learning development and model governance are becoming essential” Avner Braaverman: “What data scientists and data engineers can do with current generation serverless technologies”
Jun 9, 2019
37 min
Load more