Can We Make Model Alignment a Software Engineering Process?
The AI Conference, San Francisco, September 10, 2024, Chicago Meetups, September 25 and 26, 2024
This talk explores two unique challenges for software developers that were created by Gen AI: 1) How the non-determinism of model outputs makes testing harder, and 2) Ways we can make model tuning and related refinements more incremental and iterative.
The State of AI: A Less Technical Discussion
Private industry briefing, New York City, November 7, 2024
A discussion of the current state of generative AI: what works, what doesn't, risks, etc., prepared for a less technical audience.
Issues Blocking AI Adoption: A Decentralization Perspective
The Summit on Responsible Decentralized Intelligence, NYC, August 6, 2024
A short talk before a panel discussion on what open means in AI and one challenge blocking AI adoption, Alignment. The event was The Summit on Responsible Decentralized Intelligence —— Future of Decentralization and AI, hosted by Berkeley RDI, which aimed to bring together researchers, innovators, thought leaders, and builders to showcase and discuss responsible innovation in decentralization and AI for a greater future for humanity.
Where Is AI Headed?
Private Event, Chicago, July 22, 2024. Significant update July 25.
This talk combines and updates two AI talks below, What Issues Are Blocking AI Adoption? and AI in the Open: Why It Matters. How to Achieve It. I prepared it for a longer session at an IBM in-house event in Chicago, July 22, 2024.
What Issues Are Blocking AI Adoption?
1871 AI Innovation Summit, Chicago, June 27 2024
Despite the promise of AI, several challenges block many deployments. What do we do about them?
AI in the Open: Why It Matters. How to Achieve It.
AI Camp, Chicago, February 2024
To maximize availability and safety of AI, we should follow the path of open-source software, while recognizing what is new and different.
Open Source: Science vs. Software. What's Different? What's the Same?
Scale by the Bay, November 2023
What I learned at IBM Research about the differences and similarities of open-source software (OSS) and open-source science (OSSci).
Reinforcement Learning, ChatGPT, Games, and More
GOTO Chicago, May 2023, and IBM Research, October 2023
Things move fast; an update to January's RL talk that expands the coverage of Reinforcement Learning from Human Feedback, a key element in training ChatGPT.
Reinforcement Learning with Ray RLlib (V2 for Data Day Texas)
Data Day Texas, January 2023
Updated version of my talk on Ray's powerful reinforcement learning library, RLlib. This talk discusses reinforcement learning and how to use RLlib.
Lessons Learned from 15 Years of Scala in the Wild
Several Conferences, 2021-2022
Since I joined the Scala community roughly 15 years ago, the Scala community has learned a lot to make the language more robust and easier to use effectively. I've also learned lots of lessons about effective "enterprise" programming using Scala. Finally, I see warning signs for FP's future growth.
Next Generation AI: Transitioning to the Continuous, Self-Learning Enterprise
Draft, May 2021
A talk I'm developing that looks at the state of AI/ML and what enterprises need to do to fully adopt and leverage it.
Modularity: A Retrospective
GOTO Chicago Nights, February 18, 2020 and Scala in the City, May 28, 2020
A look at what we've accomplished in making software modular and where we need to go.
Reinforcement Learning with Ray RLlib
Several Conferences, 2020-2023
Ray comes with a powerful reinforcement learning library, RLlib. This talk discusses reinforcement learning and how to use RLlib.
Ray - Scalability from a Laptop to a Cluster
YOW! Data, July, 2020, and CodeMesh, Nov., 2020
Ray is a distributed computing system that offers a concise, intuitive API, with excellent performance for distributed workloads. It emerged out of the AI community at U.C. Berkeley.
Ray for Natural Language Processing
NLP Summit, October 7, 2020
Ray is being adopted by popular NLP frameworks like spaCy and Hugging Face. I discuss the problems Ray solves for them and how Ray is being used by them.
Executive Briefing: What It Takes to Use ML in Fast Data Pipelines
Strata San Francisco, London, and NYC 2019
A briefing for managers and executives about the challenges of serving ML models in a streaming data context.
Executive Briefing: What You Need to Know about Fast Data
Strata London and NYC 2018
A briefing for managers and executives about the trends in Fast Data and how the impact on their organizations.
Streaming Microservices with Akka Streams and Kafka Streams
Strata San Jose and London 2018, Scala Days NYC 2018, Reactive Summit 2018, YOW! 2018
I discuss processing data in microservices using Akka Streams and Kafka Streams, vs. using tools like Spark and Flink.
Stream All the Things!!
Software Architecture Conference NYC 2017, Strata London and NYC 2017, Reactive Summit 2017, ScalaIO 2017, YOW! Data 2018
I discuss the emerging architecture for large-scale stream data processing, that also integrates the best of microservice architectures.
Bash and All That
GOTO Chicago 2018
A celebration of the UNIX philosophy and the tools it spawned, like bash, make, grep, ...
Scala and the JVM for Big Data: Lessons from Spark
Strata San Jose and Singapore 2016
The JVM is the standard platform for Big Data and Scala is emerging as the standard programming language for Big Data Developers, driven in part by Spark. What lessons can we draw from this picture?
Why Spark Is The Next Top (Compute) Model
Numerous Venues 2014 and 2015
Spark has emerged as the replacement for MapReduce in Hadoop applications. This talk explains why.
Data Science at Scale with Spark
GOTO Chicago 2015
Using examples, I show how to use Spark for Data Science at scale in ways that were previously not feasible with other tools.
The Unreasonable Effectiveness of Scala for Big Data
Why Scala has proven so effective as the general-purpose programming language for Big Data development.
Why Scala Is Taking Over the Big Data World
Scala Days 2014, Scala eXchange 2014, and Data Day Texas 2015
Scala has emerged as the de facto language for big data development, driven in part by tools like Scalding and Spark.
The Internet Was Made for Cats
Chicago Scala Users Group, Jan 2016
My informal introduction to the Typelevel Cats project, including why I think it's model for open-source development.
Copious Data, the Killer App for Functional Programming
LambdaJam Chicago, others
I argued that "Copious" Data (okay, Big Data) is driving adoption of Functional Programming (FP), more so than multicore concurrency concerns, because more developers will grapple with data problems than concurrency. Because FP is based on Mathematics, it is a natural fit for working with Data, whereas languages like Java, in which Hadoop is written, are poor choices. I revisited this subject in March 2020. The original talks where presented at events in 2013-2014
SQL Strikes Back! Recent Trends in Data Persistence and Analysis
Relational databases fell out of fashion with the rise of NoSQL and Hadoop. But SQL proved too useful for too many people, so there are now many SQL-based query tools for Hadoop and subsets of SQL on several "NoSQL" databases. This talk discusses this trend and why it started. (November 4th, 2014)
The Seductions of Scala
Various Venues
An introduction to Scala that I often give at conferences and user groups. The PDF includes a lot of extra material that won't fit into a 50-60 minute time slot. The GitHub page for this talk also has the sources used for the examples. In particular, for the Akka-based Actor example at the end of the talk, see README.md. (November 19, 2013)
MapReduce and Its Discontents
QCon NYC 2012, and Big Data Techcon Boston 2013
My first public talk where I claimed that MapReduce is the Enterprise JavaBeans of our time. I criticized the MapReduce programming model and the technical limitations of the Hadoop implementation, in particular. In part, I argued that Java (pre Java-8 especially) is the wrong tool for developing Big Data applications and middleware. Instead, we should be using Functional Programming, since when we work with data, we are really doing Mathematics! (April 11, 2013)
Why Big Data Needs to Be Functional
NE Scala Symposium 2011
A more general version of the previous "Discontents" talk, where I argue that the Hadoop community needs to drop reliance on Java-centric, Object-Oriented approaches and embrace Functional Programming and languages like Scala. (April 15th, 2012)
Heresies and Dogmas in Software Development
I look at 5 ideas in the history of software development that were once popular, and still are in some quarters, but are now seen by most people as obsolete. (November 9th, 2011)
Better Programming Through Functional Programming
A half-day tutorial that introduces Functional Programming, why it has become important for our time, and how you can apply its ideas in almost any language. Examples are given in Java and Ruby. There is also a shorter talk version. (July 31st, 2011)
Polyglot and Poly-paradigm Programming
QCon San Francicso 2008
An argument that modern development problems benefit from a multi-paradigm and/or multi-language solution strategy. Different strategies are discussed in the contexts of example problems (April 2, 2011)
Hive - SQL for Hadoop
Chicago Hadoop Users Group
This talk introduces Hive, the original SQL tool for Hadoop and explains why it's a key technology that drove adoption of the ecosystem, primarily because it makes it easier to transition SQL-based data warehouses to Hadoop and it enables conventional data analysts to work with Hadoop. (January 2012)
Research Papers and Other Presentations
In addition to the talks above, here are a few other talks and the research papers I have written or co-written.Akka
- The Akka Framework: An overview of the Akka Framework for building robust, highly concurrent servers in Java or Scala.
Reactive Programming
The Reactive Manifesto lays out a vision for Reactive Programming. These talks explore various aspects of Reactive.
- Reactive Systems - The Why and the What
and Video. - Reactive Design, Languages and Paradigms and Video.
- Error Handling in Reactive Systems .
Aquarium
- Aquarium: Aspect-Oriented Programming for Ruby . There are a few exercises that go with the talk: Aquarium_RubyAOP_exercises.zip
- Aquarium: AOP for Ruby . A 30-minute talk presented at Aspect-Oriented Software Development 2008. The talk was based on my Industry Track paper.
General AOSD
- Noninvasiveness and Aspect-Oriented Design: Lessons from Object-Oriented Design Principles. Research Paper discussing classic OO design principles in the context of AOSD.
- AOP in Academia and Industry .
- Aspect-Oriented Design Principles: Lessons from Object-Oriented Design .
- AOP@Work: Component Design with Contract4J (IBM developerWorks).
- The Challenges of Writing Reusable and Portable Aspects in AspectJ: Lessons from Contract4J .
- Contract4J for Design by Contract in Java: Design Pattern-Like Protocols and Aspect Interfaces
- The Future of AOP
Other
- Accelerating automation of digital health applications via cloud native approach. Experience report of lessons learned while building the digital health application portfolio on IBM’s Accelerated Discovery Platform.