Posted in: AI, IBM Research-Australia, Publications

IBM Research at IJCAI: Pushing new AI frontiers from Down Under

For the first time in 26 years, Australia is hosting the globally revered International Joint Conference for Artificial Intelligence (IJCAI). The greatest minds in artificial intelligence (AI) from across the globe will descend on Melbourne next week to present the latest advancements in training machines to learn, reason and interact in a natural, human-like way.  It is a highly competitive conference with only 26 percent of papers accepted in 2017. This year, IBM is presenting 13 papers representing the latest research on the topic.

But before jumping into what’s happening at the next frontier of AI, let’s look at where this all started. IJCAI itself was first held in 1969, which IBM also presented at, and if you look at the proceedings, it focused on exploring the earliest milestones of areas we still investigate today – natural language analysis, pattern recognition, and machine learning. In fact, the idea of AI stems back further with the notion of creating a system which amplifies people’s own knowledge and understanding, which was presented in an essay by Vannevar Bush in 1945. Whilst today AI conjures thoughts of supercomputer intelligence systems and robots changing the world and our lives, my research peers and I know full well that the reality of AI must be taken back to that first concept of human and machine working together – augmented intelligence, if you will.

Today, some of these ideas are well and truly on our doorstep. We have taught machines to read, to see, to hear, to speak, to interact and to learn like humans do. It is an industry predicted to be worth more than $47 billion by 2020. And we’re just at the start. Next year, Gartner predicts that 30 percent of our interactions with technology will be through conversations with smart machines. IBM Research is at the forefront of building the underlying core AI technologies needed to power AI systems – as evidenced by our recent performance breakthrough and paper on distributed deep learning software, which will make it easier for enterprises to utilize machine learning and AI in a time efficient and simple manner.

So what’s happening at IJCAI to be excited about?

Below is a snapshot of IBM’s activities with a special look at two important papers:


Improving classification accuracy of deep neural networks for spiking neuromorphic chips

Antonio Jose Jimeno Yepes, Jianbin Tang, Benjamin Mashford

New learning algorithms based on deep learning match human performance, e.g. at image classification tasks and in playing strategy games. On the other hand, these new complex algorithms run on very power hungry computer architectures that prevent them from being used outside of computing clusters, thus they are not available in wearable devices (e.g. in your mobile phone) or autonomous systems. The team used a brain-inspired spiking neuromorphic computing architecture consuming power in the order of milliwatts (in comparison to hundreds or thousands of watts of traditional systems) that supports real-time processing of streaming data.  The IBM team has developed a new learning algorithm that learns a deep neural network configuration compatible with the chip and overcomes some of the shortcomings of previous algorithms. Our algorithm shows a low power solution with state-of-the- art classification performance in electroencephalogram data and the MNIST handwritten data set, which is relevant to wearable devices and autonomous systems. We think there is no better way to showcase the impact of an idea than by demoing it live, so we integrated the chip with a humanoid NAO robot by Softbank and gave it the ability to recognize objects in real-time using deep learning technology. Come to the IBM exhibitor booth and take part in our interactive demo.

Learning Feature Engineering for Classification

Fatemeh Nargesian, Horst Samulowitz, Udayan Khurana, Elias Khalil, Deepak Turaga

The Learning Feature Engineering technique removes the “hunch” in predictive modelling by automating the core task of feature engineering. It helps a data scientist reduce the time and effort required in analyses, and is poised to make data science easier and better. A team of IBM scientists has developed a novel meta-learning technique, which means learning to perform machine learning. Listen in to the talk we will deliver and learn more about this breakthrough in putting the ‘machine’ in machine learning.

Stay tuned for a more detailed blog on this innovation next week.

Other papers

Contextual Bandit with Restricted Context

Djallel Bouneffouf, Irina Rish, Raphael Feraud, Guillermo Cecchi

Neurogenesis-Inspired Dictionary Learning: Online Model Adaption in a Changing World

Sahil Garg, Irina Rish, Guillermo Cecchi, Aurelie Lozano

A Functional Dynamic Boltzmann Machine

Hiroshi Kajino

Blue Skies: A Methodology for Data-Driven Clear Sky Modelling

Kartik Palani, Ramachandra Kota, Amar Prakash Azad, Vijay Arya

Inverted Bilingual Topic Models for Lexicon Extraction from Non-parallel Data

Tengfei Ma, Tetsuya Nasukawa

A Formal Model and Mechanism for Online Organ Matching

Nicholas Mattei, Abdallah Saffidine, Toby Walsh

Efficient Optimal Search under Expensive Edge Cost Computation

Masataro Asai, Akihiro Kishimoto, Adi Botea, Radu Marinescu, Elizabeth M. Daly, Spyros Kotoulas

A Scalable Approach to Chasing Multiple Moving Targets with Multiple Agents

Fan Xie, Adi Botea, Akihiro Kishimoto

Online Bridged Pruning for Real-Time Search with Arbitrary Lookaheads

Carlos Hernandez, Adi Botea, Jorge Baier, Vadim Bulitko

Bidirectional learning for time-series models with hidden units

Takayuki Osogami, Hiroshi Kajino, Taro Sekiyama

A Notion of Distance Between CP-nets

Andrea Loreggia, Nicholas Mattei, Francesca Rossi, Kristen Brent Venable


Workshop on AI for Internet of Things – Saturday Aug 19

Anika Schumann, Cognitive Computing Scientist, IBM Research

CP & AI Workshop – Monday August 21

Panel on How Should CP be Treated in AI books and AI in CP books – Francesca Rossi, Distinguished Research Staff Member, IBM Research — Cognitive Computing

Panel on AI/CP: Applications, Industry, and Society – Francesca Rossi, Distinguished Research Staff Member — Cognitive Computing, IBM Research

Doctoral Consortium – Monday August 21

Career Panel – Nicholas Mattei, Cognitive Computing Scientist, IBM Research

Women’s Luncheon – Tuesday August 22

Joanna L. Batstone, Ph.D., Vice President and Lab Director, IBM Research – Australia

CTO IBM Australia & New Zealand

The Future of Work Panel – Wednesday August 23            

Joanna L. Batstone, Ph.D., Vice President and Lab Director, IBM Research – Australia

CTO IBM Australia & New Zealand

IJCAI AI Festival

Creative Machines – AI Lounge – Friday August 25

John Smith, IBM Fellow, Manager Multimedia and Vision, IBM Research 

Opening Plenary Talk – Industry Day – Friday August 25

John Smith, IBM Fellow, Manager Multimedia and Vision, IBM Research  

Industry Panel – Industry Day – Friday August 25

John Smith, IBM Fellow, Manager Multimedia and Vision, IBM Research

Stefan Harrer, Brain-Inspired Computing Research Scientist, IBM Research Australia

Workshops & Tutorials

Energy-based machine learning

Sakyasingha Dasgupta, Takayuki Osogami

Workshop on human interpretability in machine learning (WHI)

Kush Varshney

Demos – Robotics Showcase

Ultra-low power autonomous image classification on a NAO robot by Softbank using deep-learning and IBM’s neuromorphic chip

Hidemasa Muta, Jianbin Tang, Stefan Harrer

Controlling a bionic arm through thought and assistive visual analytics

Umar Asif, Jianbin Tang, Benjamin Mashford, Isabell Kiral-Kornek, Shivy Yohanandan, Stefan Harrer


Add Comment

Your email address will not be published. Required fields are marked *

Stefan Harrer

Brain-Inspired Computing Research Scientist, IBM Research Australia