University of Washington-Tacoma and IBM webinar updates

On May 8th-2020,  University of Washington-Tacoma and IBM Cognitive  Distinguished Engineer presented as part of OpenPOWER Academia and Research Webinar Series where 700+ folks registered and 200+ participated .

Prof Arghya Das from University of Wisconsin , Clarisse Hedglin from IBM , Prof Abhinandan from NIE India were the panelists as part of the Webinar

Dr. Linton Ward from IBM presented Covid-19 Response Capability with IBM Power Systems . He explained about IBM HPDA accelerates the medical research tasks ranging from biomarker detection, drug discovery, image restoration and classification, data fusion to quality control. In summary, he talked about IBM HPDA accelerates the applications in Genomics , Molecular Simulation , Biomolecular Structure , Diagnostics , Data Fusion and Quality Infusion

He also said about the acceleration is non-trivial due to a large amount of data, heterogeneous workflows, and models, cost, adoption, and performance issues.  The reference architecture of IBM HPDA is as shown in the figure:

He also shared some of the criticla solutions of HDPA are

Dr Linton also shared some of the customers how IBM HDPA helping key customer use cases.

Sidra medicine center, a well-known medical and research organization in Qatar, is using IBM HDPA. Due to IBM HDPA, they have reduced the time-to-completion in precision medicine. IBM IC922 has significantly improved the performance of the Cryogenic Electron microscope compared to x86 architectures. 

The IBM solutions provide users to rapidly solve the SARS-CoV-2 virus molecule structures and drug discovery to stop COVID-19 spread. NAMD and VMD simulation of influenza and COVID-19 on POWER architecture with NVLink has achieved around 27 times speedup with a runtime of 0.130s. 

Slide deck as part of the presentation

Our second speaker Prof Athirai A Irissappane , University of Washington-TACOMA presented about Leveraging GANs for Semi-Supervised Text Classification

Generative Adversarial Networks (GANs) is a class of machine learning framework where two neural network content with each other. The algorithm learns to generate new data with the same statistical properties of the training set.  In many domains (such as opinion spam classification), obtaining ground truth labels is often time-consuming and costly. Deep generative models have shown promising results for semi-supervised learning. Specifically, GANs can generate samples very close to real data have achieved state-of-the-art results. However, most research on GANs is for images (continuous values) and not text data (discrete values).  

As part of the presentation , She was showing how GANs could be leveraged for text classification. Also, the state-of-the-art language models such as GPT-2, a language model with 1.5 billion parameters were introduced and can be integrated for better classification accuracy. She has got special access to IBM AC922 POWER9 systems to run her experiements.

Recorded sessions can be viewed from our youtube channel - OpenPOWER AI virtual University

These two talks well received by audience and there were several questions about the capabilites of IBM POWER9 systems as well as the solutions being deployed using various features of IBM Cognitive stack.

I would like to thank the panelists , presenters and participants for their continuos support on the OpenPOWER Webinar series .