In this post, we’re going to look at how advanced deep learning technology can help a university to strengthen its partnerships and industry links. And we’re going to do so by sharing a success story.
At Queen Mary University London (QMUL), two IBM Power systems S822LC systems supplied by us have led to all manner of innovations. Codenamed ‘Minsky’, these powerful deep learning machines have also supported the establishment of valuable private sector links.
Recently, QMUL found itself celebrating the award success of the commercial CCTV solutions provider SeeQuestor. In March 2017, SeeQuestor won the 12th annual ADS Security Innovation Award for its time-saving video intelligence toolkit.
But what does this have to do with QMUL, OCF and the importance of partnership? Let’s find out.
‘Minsky’ quickens Vision Research insights
The Vision Lab’s research is truly at the cutting edge. Often funded by Government agencies such as the Ministry of Defence, the Home Office, and the Royal Society, it has also partnered with industry leaders such as BAE Systems, BAA, BT Labs, and BBC R&D.
So, when OCF helped Queen Mary become one of the first British universities to acquire a Power System S822LC – ‘Minsky’ – for research use, Professor Gong and his team soon put it to work.
The lab used ‘Minsky’ in its projects to create new video analytics technology, which can learn about human behaviour and activity in video data without supervision.
This self-learning can reduce the need for human CCTV operators, saving thousands of work hours.
The research involved analysis of massive volumes of video data. But findings were reached much sooner thanks to ‘Minsky’, which features two POWER8 CPUs and four Tesla P100 GPUs, plus super-fast NVLINK processor inter-connects that help deep learning applications run much, much faster.
Against a comparable x86-based system, ‘Minsky’ can deliver more than twice the performance including:
- 5x more Kinetica “Filter by geographic area” queries per hour
- 9x more GFLOPS while running LatticeQCD
- 2x more “Base Pairs Aligned” per Second running SOAP3-dp with 2 instances per device
The Vision Lab’s intelligent video analysis technologies were so impressive, QMUL decided to patent the technology and launch a commercial spin-out: Vision Semantics Limited (VSL).
Two of the key abilities of VSL’s technology are:
- Multi-camera tracking – Auto-tracking the activity of one individual across multiple video sources, such as an array of cameras in a shopping mall.
- Crowd analysis – Analysing crowd dynamics and behaviour in public spaces, including estimating crowd density, profiling crowd distribution over space and time, and analysing demographic distribution.
These advanced features have led to industry partnerships between VSL and EPSRC, the Metropolitan Police, QinetiQ, and other leading organisations.
Beating the competition
VSL’s technology is also fundamental to SeeQuestor video intelligence solutions, which brings us back to the beginning of our story.
The company beat well-established competitors to the 2017 ADS Security Innovation Award. And its success all began with Professor Gang, the Computer Vision Laboratory, and ‘Minsky’s remarkable deep learning capabilities.
Speaking at the award ceremony, Professor Gang said:
“The commercial system built on Vision Semantics’ patented video analytics technology beat some world-leading products, including QinetiQ’s latest SPO-NX system built for the US Transport Security Administration.
“This consolidates the rapidly gained reputation of this video analytics technology among end-users in the US, UK, Australia and India.”
What could ‘Minsky’ do for your research?
If you’d like to know more about the Power System S822LC, and how deep learning could enhance your own research projects, please get in touch. Leave a comment below or find me on Twitter @maheshpancholi.