We currently focus on automating and accelerating AI/SW/HW co-design to enable efficient, self-optimizing and self-learning systems. We use Collective Knowledge framework and repository to crowdsource AI/SW/HW autotuning in terms of speed, accuracy, energy, memory, size and costs across diverse hardware, environments and realistic data sets, and combine it with machine learning and run-time adaptation (DATE'16, CPC'15, JSP'14. ):
In the past, dividiti founders participated in a number of highly successful research international projects including MILEPOST which produced machine learning based self-optimizing compiler as well as a public repository of optimization knowledge considered by IBM to be the first in the world!
In 2017 we received a test of time award for the ACM CGO'07 research paper that led to creating Collective Knowledge - this annual award recognizes outstanding papers published at the ACM/IEEE International Symposium on Code Generation and Optimization (CGO) one decade earlier, whose influence is still strong today!
Publications from our team: