(copy of a post appeared on BSC-CNS blog)BSC-positionpaper

Our last newsletter issue presents the BSC view about the challenges of Cognitive Computing. Some of you asked us a more extensive explanation. Here we go!

Cognitive Computing?

Intelligent systems are mainstream now in most industries [1]. They use artificial intelligence, natural language processing, machine learning, artificial neural networks, streaming analytics, unstructured data and Internet of Things, among many other techniques and technologies.

Many of the technologies and events of the every day’s life are directly or indirectly influenced by these intelligent systems. Furthermore, the use of these intelligent systems will impact our society massively: new medical models for personalized medicine, cars with cruise control, drones or robots of all types, etc.

We are at the dawn of a new age of computing where systems will learn at scale, reason with purpose, and interact with humans naturally [2]. These systems will not be programmed; they will be trained to sense, predict, infer and, in some ways, think, hence the name of Cognitive Computing. This will be achieved using artificial intelligence and machine learning algorithms that are exposed to massive amount of data [3].

Available to everyone

These systems will improve over time as they will build knowledge and acquire depth in specialty domains. Many of them are available to everyone, like the case of Watson Computer [4] from IBM, a text-based system capable of answering question through natural language processing, or the recent advances in computer vision with the broad adoption of deep learning, that has opened a whole new era with exciting results as human skills in tasks such as image classification or person verification. This is the case of Google Cloud Vision API [5] , released some weeks ago by Google.

Challenges

We are entering into an extremely exciting period for interdisciplinary research required for the advancement of Cognitive Computing. Their challenges can be summarized in three main pillars:

  • HPC & Big Data Technologies
  • Scientific Applications
  • Advanced Analytics

Pillar 1: New advancement on HPC and Big Data Technologies

The first real breakthrough for cognitive computing came when parallel processing was developed, like IBM Watson. However, these reasoning capabilities, the complexity of the data, and the time to value expectations are driving the need for a new class of supercomputer systems. A lot of research is still necessary in order to integrate the best analytics knowledge with new Big Data technologies and the awesome power of emerging computational systems in order to interpret massive amounts of heterogeneous data at an unprecedented rate.

Pillar 2: Cognitive Computing as a smart assistant for scientific applications

Cognitive Computing will enable new research models, and change the way that the entire research work is done. It combines massive data sets with sophisticated analytics to help human experts synthesize findings and improve decision-making. Cognitive tools are intended to work with human scientists, amplifying and increasing their intellect, as part of collaborative human-machine systems. It will open up new modes for actually doing science, as the just released Accelerating Science: A Computing Research Agenda [6] pointed out about the new role for Cognitive Computing in Science.

Pillar 3: Advanced Analytics

Many challenges lie ahead and we think that the community will focus their efforts not only on natural language or images processed by the next generation systems, but also on multimedia unstructured data with unsupervised learning (including media streams, IoT sensor data, and other non-traditional sources).  Unsupervised learning remains a major challenge (in opposite to supervised learning, where the software is trained with data labelled by humans). We are likely to see increasing activity in this arena, especially around unstructured data that continues to grow exponentially. But also in scenarios where the amount of data available are reduced, we will see new research opportunities building realistic intelligent systems that mimic how humans learn, which requires a very small amount of training examples and it is almost entirely unsupervised [7].

Conclusions

Undoubtedly the Cognitive Computing requirements are driving the need for a new class of supercomputer systems. This is the reason why many supercomputing centres in the world are playing a big role in the advancement of Cognitive Computing.  These centres are gradually adding to their continuous contributions to improve the supercomputing architectures and runtimes environments support to powerful middleware’s for big data workloads. More recently, there are centres that contribute to achieve a High-Performance Big-Data Analytics.

Scientific progress is typically the result of an interdisciplinary, long and sustained effort by a large community rather than a breakthrough, and cognitive computing is not an exception. We are entering into an extremely exciting period for interdisciplinary research, where ecosystems like the ones from BSC-CNS, will evolve to cope with new challenges.

Jordi Torres, Research Manager & Director’s Senior Advisor at BSC  –  Mateo Valero, BSC Director 

 

[1] “La computación Cognitiva en el diario El País”. Jordi Torres ‘s Blog. October 2014. [Online]. Available at: http://jorditorres.org/la-computacion-cognitiva-en-el-diario-el-pais [Accessed 27/02/2016]

[2] Dan Briody, “New Vocabulary: Cognitive Computing,” THINK Leaders, October 2015. [Online]. Available at: https://www.think-exchange.com/how-to/cognitive-computing-new-vocabulary [Accessed 27/02/2016]

[3] World Economic Forum Annual Meeting, Davos, January 2016 [Online]. Available at: http://www.weforum.org/reports/the-future-of-jobs [Acessed: 27/02/2016].

[4] Watson, the platform for cognitive business, IBM. [Online]. Available at: http://www.ibm.com/smarterplanet/us/en/ibmwatson/ [Accessed 27/02/2016]

[5] Google Cloud Vision API. [Online]. Available at: https://cloud.google.com/vision/ [Accessed 27/02/2016]

[6] “Accelerating Science: A Computing Research Agenda”, Vasant G. Honavar, Mark D. Hill, and Katherine Yelick. Computing Community Consortium . 19/02/2016 . Available at: http://cra.org/ccc/wp-content/uploads/sites/2/2016/02/Accelerating-Science-Whitepaper-CCC-Final2.pdf [Accessed 27/02/2016]

[7] Lake, Brenden M., Ruslan Salakhutdinov, and Joshua B. Tenenbaum. “Cognitive Science: Human-level concept learning through probabilistic program induction.” Science 350.6266 (2015): 1332-1338. Available at: http://web.mit.edu/cocosci/Papers/Science-2015-Lake-1332-8.pdf [Accessed 27/02/2016]