#LAK19 Conference Report [Part 2]: Main Conference

#LAK19 Conference Report [Part 2]: Main Conference

(This is the second part of a report of the #LAK19 conference. See Part 1 here.)

It was the only beginning of many fun conversations when the Main Conference kicked off on Wednesday!

Our conference and program chairs delivered an exciting welcome message.

LAK is becoming larger,

but still selective – with a 33% acceptance rate for Full and Short papers.

LAK is international – given representation and collaboration patterns. And interdisciplinary – based on which publication sources we cite.

See their slides for more detailed information.

Keynotes 1 and 2

With a young family, I could only stay till Thursday and had to miss the third keynote by Shirley Alexander (Professor & Deputy Vice-Chancellor, University of Technology Sydney, Australia) on how to nurture a Data-Intensive University.

The other two keynotes were delivered by Ryan Baker (UPenn) and Lise Getoor (UC Santa Cruz). I am reviewing these two keynotes together because they were both slightly more technical (compared by the third keynote by Alexander) and both about addressing challenges in building models.

Ryan Baker is the founding president of the international society of Educational Data Mining (EDM), a sister community of learning analytics. His talk was focused on six important challenges faced by both EDM and learning analytics. As a keynote it almost sets a future research agenda for the community.

Predictive models has been a trending topic in the community even though prediction is only one of those five EDM techniques outlined by Baker and Yacef (2009) (see also Siemens, 2013).

I appreciated Ryan’s keynote because it pointed out a few acute problems with our current research and practice. We build models and analytics facing a serious constraint – the ‘learning system wall’ problem (Challenge #1). Learning analytics systems we build are not making lasting impact on student lives (Challenge #2). Findings from models are not necessarily interpretable (Challenge #3). We are less confident in detecting student learning in real-world scenarios (say, not in front of a computer) (Challenge #4). Our models are not generalizable across contexts (Challenge #5). Our models are not general either – not built for a non-typical population (e.g. high school dropouts, Alaska students) (Challenge #6).

These important challenges are closely aligned with LAK19’s theme, Promoting Inclusion and Access, and vital for the long-term legitimacy of this field. Framing them in technical terms was intended to help us find practical solutions. But I started to feel less optimistic about addressing them within the next 18 years when considering the complex political, ethical, socio-technical, theoretical terrain facing these challenges.

(Note: Expect Ryan to share more details about these challenges via Twitter soon.)

The second keynote by Prof. Lise Getoor was focused on Collective Reasoning as a key challenge in data mining and machine learning. She first called attention to structures in data and critiqued the widespread practice of flattening rich behavioral data into tables. Starting from this recognition, she highlighted that all decisions are structured and argued we should use multimodal, multi-relational, spatio-temporal features of data when making decisions, instead of flattening data and discarding relations.

Getoor went further to explain Collective Reasoning and suggested constructing logial rules – many of them – to further support reasoning.

Her group developed the Probabilistic soft logic (PSL) framework and toolkit for building probabilistic models. She then went ahead to introduce the application of PSL in a wide range of problem spaces (including MOOC discussion forums).

Issues discussed by Getoor spoke to my heart. In the learning sciences, we’ve been criticizing the practice of coding-and-counting approach. While part of my research has been trying to cast light on micro-temporal dynamics in learning processes, I am also interested in maintaining the complexity of learning data intact when conducting data analysis. The keynote itself did not provide more details about Probabilistic Soft Logic (PSL), given the time limit, so I look forward to digging into the PSL framework and tools later.

Learning Sciences Panel

On Wednesday, I moderated a Learning Sciences panel featuring Ryan Baker, Nancy Law (University of Hong Kong), Phil Winne (Simon Fraser University), and Alyssa Wise (New York University). We were charged to inspire the audience with “Design and analysis strategies that promote inclusion and increase access”. While all of our panelists had very wise things to say, we all concurred we need more expertise represented in the room.

Near the end of the session, one of this year’s Doctoral Consortium participants Catherine Manly (UM-Amherst) mentioned her work on Universal Design for Learning. We need more work like this!

Sessions on Text & Discourse, Design, Ethics

The rest of my conference participation was a bit disjointed. Besides my own presentation on Value Sensitive Design, I was only able to attend a few sessions on Ethics (which I chaired), Dialogue & Engagement, and Reading Analytics.

There were many sessions I wished to attend. And I missed an important talk on the impact of learning analytics, by Shane Dowson and colleagues.

Fortunately, the conference proceedings are nicely organized by sessions for us to catch up.

Slides of my talk on Towards Value-Sensitive Learning Analytics Design can be found below. I am excited about this direction!

LAK19 - Towards Value-Sensitive Learning Analytics Design

Looking Ahead

It has been a great LAK!

Given it will be LAK’s 10th anniversary, I do wonder how much this community has achieved and what needs to be down in the next 5 or 10 years.

Below I offer my own list of challenges:

  1. Work with the larger system. While we’re doing detailed, careful work, it is important to make sure our work is contributing to the betterment of the larger education system. We need to connect models with design, and with feedback. We need to engage more voices in our work. We need to look at choice-making by learners besides knowledge acquisition. We need to connect learning that’s life-wide and lifelong.
  2. Move towards knowledge convergence. While we can do all sorts of cool analyses on all sorts of interesting data, we need to be an even better job on understanding learning in richer contexts. How can we put advances from multiple analytical perspectives together?
  3. Focus on impact. While there are convincing cases of data-rich educational practices, we need to invest more in studying acceptance and impact.
  4. Include one keynote speaker from a “non-WEIRD” country each year. While the society has been doing an amazing job on equity and inclusion, I would like to see a stronger representation from the Global South in the next 10 years. Ensuring voices from “non-WEIRD” countries could be a good start.

Look forward to seeing you at LAK ‘20 in Frankfurt, Germany.

Avatar
Bodong Chen
Associate Professor of Learning Technologies

Associate Professor in learning technologies at the University of Minnesota.