A Review of AAAI-20

The 34th annual AAAI Conference on Artificial Intelligence conference was held at Hilton New York Midtown Hotel in New York, USA during February 7–12, 2020. As a continued tradition, the 32nd annual conference on Innovative Application of Artificial Intelligence (IAAI) and 10th Symposium on Educational Advances in Artificial Intelligence (EAAI) were also co-held with the AAAI conference at the same venue.

When it comes to the top-notch forums for any researcher/practitioner to communicate about their latest and greatest innovations in the field of AI, AAAI conference undoubtedly falls in that bracket. Given a long legacy and a reputation for maintaining the highest quality review standards, AAAI is widely considered one of the most prestigious conferences in the field of AI. This is also reflected in the sheer growth trends of this conference. AAAI-20 conference witnessed 8800 paper submission (24% up than the previous year), 1591 accepted papers (20.6% acceptance rate, 4.4% higher than the previous year). As far as attendance is concerned, this year was somewhat unlucky due to the Coronavirus. The effect of Coronavirus was noticeable in the attendance of AAAI-20. A large number of authors and participants from China could not attend this year’s conference. While the program committee worked hard to arrange for video recordings of such speakers at the last minute but still we could feel a void — there is probably no substitute for having face-to-face interaction, no matter how much AI technologies get advanced!!

This year, the conference also experimented with a mobile app called “Guidebook” to make the schedule of the conference, workshops, and tutorials available to the participants on their fingertips for an easy switchover across parallel sessions.

Major Themes @ AAAI-20

The schedule of AAAI main conference for this year included papers across pretty much the entire spectrum of AI. However, certain topics/themes were clearly dominating this year’s program.

An extraordinarily large number of papers were presented in the broad field of NLP with special focus on entity/relation identification and linking, machine translation, text generation, dialogue, text classification, machine comprehension, question answering, etc. It is worth mentioning that neural language models such as BERT and Transformer were omnipresent in most of these thematic sessions. These deep-net based language models appeared to be the talk of the town and seemed to have become darling of NLP researchers in such a short span since their inception.

Another prominent area was related to computer vision and image understanding. This included multiple sessions on themes such as visual question answering, image retrieval, segmentation, ranking, scene understanding, domain adaptation, transfer learning, language grounding, etc. Lastly, the classical areas of AI also had their fair share including knowledge representation, reasoning, planning, CSP, optimization, reinforcement learning, logic, game theory and mechanism design, graphical models, etc.

Keynotes, Invited Talks, Special Events

Among many highlights, the following three events probably attracted the maximum number of audiences — (i) ACM Turing 2018 Award Winner Event and Panel, (ii) AI History Panel: Advancing AI by Playing Games, (iii) Oxford-style debate.

The first event included three legendary figures in the field of deep learning, namely Yoshua Bengio, Geoffrey Hinton, and Yann LeCun. Each one of them delivered a fantastic talk about the past, present, and future of deep learning/ai technologies.

The AI History Panel had a fantastic set of panelists including Murray Campbell (IBM), Michael Bowling (University of Alberta), Hiroaki Kitano (Sony), David Silver (DeepMind and University College London), and Garry Kasparov (Russian Chess Grandmaster). All of the panel members have made a significant contribution to realizing the AI systems that can play games against humans. This includes IBM Deep Blue, Deep Mind’s AlphaGo, etc. The panelist shared their perspective about ingredients that went into developing such technologies and the challenges that they encountered. Undoubtedly, this was an inspirational session for the newbies in the field of AI.

At the end of the very hectic day, the oxford-style debate was deliberately designed to be a very entertaining event yet thought-provoking event. The topic for the debate was “Academic AI researchers should focus their attention on research problems that are not of immediate interest to industry”. Both the teams (for and against) put forward very interesting arguments. The audience really had great fun watching this event live.

Other highlights of AAAI-20 includes a fireside chat with Nobel Laureate Daniel Kahneman — author of the famous book “Thinking Fast and Slow”. The chat was centered around the present and future of AI and human decision making.

Last but not the least, there was a very impressive keynote talk by one more AI legendary, namely Stuart Russell, who is well known for his famous book on “Artificial Intelligence: A Modern Approach ”.

The IAAI-20 program also had some very exciting keynote speeches. This year’s Robert S. Engelmore Memorial Award was sponsored by IAAI and was conferred on Henry Kautz. As part of this award lecture, Kautz talked about the history of AI covering three summers and two winters of AI. Prof. Kautz also argued why there might not be any third winter for AI. Another interesting IAAI keynote talk was delivered by David Cox who is leading MIT-Watson AI Lab from the IBM side. David gave a fantastic overview of the Neuro-Symbolic program and various programs that his team has made in recent times while collaborating with Prof. Joshua Tenenbaum from MIT.

Awards

Several award-winning works were presented throughout the conference.

The AAAI-20 outstanding paper award was conferred on the paper “WinoGrande: An Adversarial Winograd Schema Challenge at Scale”, by Keisuke Sakaguchi, Ronan Le Bras, Chandra Bhagavatula, Yejin Choi. This paper kind of highlighted limitation (human biases) of the well-known Winograd Schema Challenge which was proposed as an alternative to the Turing Test in the year 2011.

This year’s outstanding student paper award went to “Fair Division of Mixed Divisible and Indivisible Goods” by Xiaohui Bei, Zihao Li, Jinyan Liu, Shengxin Liu, Xinhang Lu.

The outstanding paper award under special track on AI for social impact was conferred on “A Distributed Multi-Sensor Machine Learning Approach to Earthquake Early Warning”, by Kevin Fauvel, Daniel Balouek-Thomert, Diego Melgar.

Also, there were honorable mentions in each of the above categories as well as Blue Sky Ideas awards.

The classic paper award was conferred on “QUICKXPLAIN: Preferred Explanations and Relaxations for Over-Constrained Problems” by Ulrich Junker.

Finally, there were 10 papers presented in the IAAI-20 conference which received deployment application awards.

Tutorials and Workshops

Tutorials and workshops were organized before the start of the main conference. There were a couple of tutorials which pulled heavy crowd due to the contemporary nature of the topic. This list includes tutorials on (i) Graph Neural Networks: Models and Applications, (ii) Explainable AI: Foundations, Industrial Applications, Practical Challenges, and Lessons Learned, (iii) Recent Advances in Transferable Representation Learning.

This year’s AAAI had a fantastic set of workshops. The ones which were standing out from the point of view attendees strength include (i) Deep Learning on Graphs: Methodologies and Applications.

Socials and Meet-ups

Different workshops, Doctoral Consortium as well as companies were inviting students and researchers for an informal meet and greet type social events and lunches for better interaction during the conference. Many companies had events with invite-only participation that was circulated over email. Google invited people for the Black in AI evening, while Apple invited relevant candidates for more networking. Bloomberg distributed invites to candidates with similar research experience to their office, while Jane Street invited for lunch. These social events give candidates the opportunity to talk with researchers on similar workstreams and recruiters to gauge possible career options.

Industries@AAAI-20

AAAI 2020 witnessed AI Job Fair on February 11, 2020, from 12:30–3:00 PM in the Rhineland Gallery, Hilton New York. It was attended by 38 companies who get 1 minute of time to present their company, current opportunities and invite people to their booths. There were several companies including Amazon, Baidu, Google, IBM Research, Mayo Clinic, Microsoft Research, Sony, Visa Research, several startups, publishing houses, and 3 universities. Several of these companies had their booths since day 1 of the conference which invited a lot of students and researchers to attend the ongoing demonstrations and discuss career opportunities. The Job Fair had an option to submit candidate resumes, months before the conference as well as at the company booths, which attracted a lot of attendees.

IBM@AAAI-20

IBM Research had a significant presence at AAAI with approximately 50 papers in the main conference alone. These papers addressed the problems and challenges in several subfields of AI including question answering, dialogue, planning, bias and fairness, language understanding and generation, and temporal modeling. The corresponding talks attracted several researchers who actively participated in these talks by asking interesting questions. A team from IBM Research was also awarded the best demo at AAAI-20 for TraceHub — a platform to bridge the gap between state-of-the-art time-series analytics and datasets.

Apart from the talks, there were several emerging technology demos that were showcased at the IBM Research booth as part of “Emerging Technology Pedestal”. These demos were showcased for three days, with each demo being showcased for 45 mins. The demos covered broad areas under AI and received a lot of interesting questions, comments, and suggestions. Some interesting demos were:

• CLAI — short for Command Line AI Toolkit, aims to bring the power of AI to the shell by augmenting the user experience with natural language support, troubleshooting, and automation, as well as providing researchers with an extensible API to develop their own AI plugins.
• GAAMA — short for Go Ahead Ask Me Anything, GAAMA is a (multi-lingual) reading comprehension system for question-answering.
• AI Automation for Visual Inspection — enables civil engineers to efficiently create and manage AI models for visual inspection of civil infrastructures like bridges.
• Data Readiness Toolkit for AI — demonstrates how AI techniques can be used for AI readiness evaluation
• COPA: a COnversational assistant for business Process Automation — a slackbot agent created for digital process automation for travel
• LAMBADA — Language Model Based Data Augmentation method for text classification tasks.
• Causal Knowledge Extraction Through Large-Scale Text Mining — showcasing pipelines for automatically extracting causal relations from the text, with application to building models for Enterprise Risk Management.
• Doc2Dial — a framework for dialogue composition grounded in documents.
• TraceHub — a platform to bridge the gap between state-of-the-art time-series analytics and datasets.

IBM India Research Lab (IRL)@AAAI-20

IBM India Research Lab had its fair share of presence at AAAI-20. Three papers in the main conference were authored by researchers at IBM India Research Lab:

· Translucent Answer Predictions in Multi-Hop Reading Comprehension

· An ADMM Based Framework for AutoML Pipeline Configuration

· Mask and Focus: Conversation Modelling by Learning Concepts

The corresponding oral presentations as well as posters were well received and attracted several researchers working in related fields in the industry as well as academia.

Another paper titled Question Quality Improvement: Deep Question Understanding for Incident Management in Technical Support Domain was presented at IAAI-20, a conference that was co-located with AAAI-20. This work has been running on production systems for more than two years now and the authors were awarded the Deployed Application Award.

Summary

In summary, we can say that despite the Coronavirus setback, this year we had a fantastic program and unforgettable experience after attending the AAAI-20 conference. The whole proceeding of the conference is available online.

Authors

This blog entry was authored by

- Anupama Ray, Research Scientist @ IBM Research — India

- Dinesh Garg, Research Scientist @ IBM Research — India

- Gaurav Pandey, Research Scientist @ IBM Research — India

- Hima Patel, Research Manager @ IBM Research — India