SEMLA 2021 Program

Tuesday, June 8, 2021

Speaker : Ben Shneiderman, Professor of Computer Science, University of Maryland

Title: Human-Centered AI: What it is and how software engineering can contribute to its success.

Video Recording: https://www.youtube.com/watch?v=7soeKLx9YsQ

Date/time: June 8, 2021, 12:00pm-1:00pm EDT

Abstract:

A new synthesis is emerging that integrates AI technologies with HCI approaches to produce Human-Centered AI (HCAI). Advocates of this new synthesis seek to amplify, augment, and enhance human abilities, so as to empower people, build their self-efficacy, support creativity, recognize responsibility, and promote social connections. These passionate advocates of HCAI are devoted to furthering human values, rights, justice, and dignity, by building reliable, safe, and trustworthy systems. The talk offers three ideas:
      • HCAI framework, which shows how it is possible to have both high levels of human control AND high levels of automation
      • Design metaphors emphasizing powerful supertools, active appliances, tele-operated devices, and information abundant displays
    • Governance structures to guide software engineering teams, safety culture lessons for managers, independent oversight to build trust, and government regulation to accelerate innovation.
The talk will emphasize the software engineering practices that will make machine learning more reliable, by increasing audit trails, reducing bias, and supporting explainability. These ideas are drawn from Ben Shneiderman’s forthcoming book (Oxford University Press, January
2022). Further information at: https://hcil.umd.edu/human-centered-ai  Join the Human-Centered AI Google Group at:  https://groups.google.com/g/human-centered-ai and follow @HumanCenteredAI on Twitter.




Tuesday, June 15, 2021

Speaker : Nisha Talagala, CEO and Co-founder of Pyxeda AI and AIClub.

Title: Software Innovation to Enable Broad-Based AI Literacy

Video Recording: https://www.youtube.com/watch?v=2BdCc3usiPA

Date/time: June 15, 2021, 12:00pm-1:00pm EDT

Abstract:

The AI market is projected to grow to $190 Billion by 2025. AI is being used in every industry and is projected to be a core skill for the future. While lack of production machine learning (MLOps) was a limiter in the last few years, these limits are starting to be overcome, with MLOps practices now standard in many organizations. More businesses are starting to see positive returns from their AI initiatives.  We are shifting to a new phase of AI development, where broad segments of the non-technical workforce are encountering AI in their job roles. This is both exciting and fraught with peril.  Instances of AI failures, legal issues, and ethical issues are rising. There is pressure on AI development to accommodate not just data scientists but people from all walks of life.  

In this talk, we discuss recent AI and Machine Learning technology trends and the role of MLOps in driving AI commercial success. We then discuss what it takes to bring AI knowledge out of the technical domain and into the broader workforce, and technology trends like low-code that enable broad adoption. We will describe a software framework that enables AI Literacy via a number of novel approaches – including automation of data preparation, automated AI compiler/code generation, agile iteration, and optimization of the entire AI lifecycle. We will then discuss the use of this software infrastructure in AWS and GCP to implement a framework for AI Literacy – the Four Cs  – and experiences of bringing AI literacy to individuals worldwide.

Thursday, June 17, 2021

Speaker : Q. Vera Liao, Research AI at IBM T.J Watson Research Center

Title:Questioning the AI: Towards Human-Centered Explainable AI (XAI)

Video Recording: https://www.youtube.com/watch?v=TT6psmj55Tc

Date/time: June 17, 2021, 12:00pm-1:00pm EDT

Abstract:

Artificial Intelligence technologies are increasingly used to make decisions and perform autonomous tasks in critical domains such as healthcare, finance, and criminal justice. The needs to understand AI in order to improve, contest, develop appropriate trust and better interact with AI systems have spurred great academic and public interest in Explainable AI (XAI). Recently, open-source toolkits, including IBM Research’s AI Explainability 360, are making a growing collection of XAI techniques into practitioners’ toolbox. My colleagues and I at IBM Research conduct human-computer interaction (HCI) research that aims to empower AI
practitioners to make effective and responsible use of such a toolbox to create good XAI user experiences. Meanwhile, our work provides insights into real-world user needs for AI explainability to inform gaps and opportunities for XAI algorithmic research. Our work follows two complementary paths. First, we conduct HCI research by designing and studying XAI systems of various use cases in the AI lifecycle. Second, we study AI design practices of product teams and engage with the design community to develop and advocate for user-centered design processes for XAI. I will conclude the talk with lessons learned for bridging the process of creating responsible AI systems and empowering people in the process.


Tuesday, June 22, 2021

Speaker : Danilo Sato and Arif Wider,  Head of Data & AI Services at ThoughtWorks UK and professor at HTW Berlin

Title: Continuous Delivery for Machine Learning

Video Recording: https://www.youtube.com/watch?v=um6Sq5EhW6A

Date/time: June 22, 2021, 12:00pm-1:00pm EDT

Abstract:

Machine Learning applications are becoming popular in our industry, however the process for developing, deploying, and continuously improving them is more complex compared to more traditional software, such as a web service or a mobile application. Continuous Delivery for Machine Learning (CD4ML) is the discipline of bringing Continuous Delivery principles and practices to Machine Learning applications. In this talk we will share our industry experiences implementing CD4ML, introduce its technical components, and explore what future challenges
need to be solved.


Thursday, June 24, 2021

Speaker : Yingnong Dang, Principal Data Scientist Manager at Microsoft Azure

Title: AIOps: From Research Innovations to Industrial Adoptions

Video Recording: https://www.youtube.com/watch?v=agetfkYpr7Y

Date/time: June 24, 2021, 12:00pm-1:00pm EDT

Abstract:

The scale and complexity of cloud computing has been ever-increasing. This brings challenges on effectively building and managing cloud computing systems that are highly efficient and reliable, enable high customer satisfaction, and achieve high engineering productivity. In this talk, I will first share an AIOps vision of infusing AI into the cloud computing platform and DevOps process. I will then share a few AIOps efforts in Microsoft Azure to demonstrate how an AIOps solution can be built and adopted in industrial settings. Specifically, I will share how Azure uses intelligent anomaly detection and correlation for safeguarding the rollouts of hundreds of component payloads to millions of machines spreading in 60+ Azure regions across five continents (project Gandalf safe deployment).

I will also share how we built a resilient mechanism for Azure against failures by employing ML-based prediction and an online learning mechanism (project Narya). I will then talk about our learnings on engineering AIOps solutions, and a few open challenges on cloud computing that need more research and innovations in the related areas including software engineering and systems.

Thursday, July 1, 2021

Speaker : Grace Lewis, Principal Researcher at CMU Software Engineering Institute

Title: Architecting ML-Enabled Systems

Video Recording: https://www.youtube.com/watch?v=X1pilVNPne0

Date/time: July 1, 2021, 12:00pm-1:00pm EDT

Abstract: 

Developing software systems that contain machine learning (ML) components requires an end-to-end perspective that considers the unique life cycle of these components — from data acquisition to model training to model deployment and evolution. While there is an understanding that ML components in the end are software components, there are some characteristics of ML components that bring challenges to software architecture and design activities, such as data-dependent behavior, drift over time, and timely capture of ground truth to inform retraining. The goal of this talk is to highlight some of these challenges, along with proposed practices and remaining gaps for successfully architecting ML-enabled systems.