The paper, entitled “Low Complexity Multiply Accumulate Unit for Weight-Sharing Convolutional Neural Networks” is a 4 page paper, the PDF for which can be found by searching arxiv.org and directly at http://arxiv.org/abs/1609.05132.
The paper, entitled “Low Complexity Multiply Accumulate Unit for Weight-Sharing Convolutional Neural Networks” is a 4 page paper, the abstract for which is as follows:
“Convolutional Neural Networks (CNNs) are one of the most successful deep machine learning technologies for processing image, voice and video data. Implementations of CNNs require very large amounts of processing capacity and data, which is problematic for low power mobile and embedded systems. Several designs for hardware accelerators have been proposed for CNNs which typically contain large numbers of Multiply Accumulate (MAC) units. One approach to reducing data sizes and memory traffic in CNN accelerators is “weight sharing”, where the full range of values in a trained CNN are clustered and the cluster index is stored instead of the original weight value. In this paper we propose a novel MAC circuit that exploits clustering in weight-sharing CNNs. Rather than computing the MAC directly we instead count the frequency of each weight and place it in a cluster. We then compute the accumulated value in a subsequent multiply phase. This allows hardware multipliers in the MAC circuit to be replaced with adders and selection logic. Experiments show that our approach results in fewer gates, smaller logic, and reduced power.”
If anyone wants to comment on the paper, send me your comments.
Well done to the Sing Aloud choir (the choir in which Zara and Naomi sing) who performed brilliantly this evening in Killeshin Church in Carlow. They really enjoyed themselves singing and we enjoyed hearing them.
I left Xilinx on 1st April 2016 (and it was no April Fool’s) after nearly 12 years of working in the DSV department in Xilinx so we went for a few beers in Neary’s in Dublin. Here are a few photos of the lads in Neary’s enjoying a Friday night pint or five.
In attendance were Semir Bajramovic, Ciaran McGloin, Stephen Ralph, Joshua Ajuebor, Michael Dovel, Amr Monawir, George Duffy, Anthony Collins, Bob Tait, Austin O’Riordan, Jonathan Hunt, John O’Dywer, Eugene O’Rourke, Wojciech Wieczorkowski, Amit Wadhwa, Juan Carlos Rodriguez Cruces and James!
After that a few of us who wanted more went with me to The Dean. It turns out I used to work in the same place when it was Parthus back in 1999-2002. It was strange to see the old place again, although the reception now had a DJ booth in it’s place and the meeting rooms were now bars. Up in the roof space (where we used to have coffee and cakes on a Friday morning) the walls had now been replaced with windows so you could see the view of Dublin. Also they’d installed swings by the lifts … Corporate places should really have more swings.
It was a great working for Xilinx for the last 12 years and have had some great times with an excellent team. I was great to see everyone out, we were all in great form and what a great way to end my present period in industry and mark the start of my academic career.
Now it’s time to focus on my PhD with a bit of part time lecturing!
Just a short post today to state that today is the first day of my PhD Research degree! I’ll be working in the field of Deep Learning, studying at Trinity College Dublin.
PhD Research Area
I’ll be researching how deep learning neural networks might be implemented in a low power way in FPGA’s. Dr. David Gregg, Associated Professor of the School of Computer Science and Statistics, will be my supervisor.