Most organizations spend significant resources on their cloud computing bill or their internal cluster’s operational costs. Moreover, interactive-speed analytics and real-time machine learning have remained elusive goals even for small datasets, due to a large number of users sharing the same infrastructure, communication bottlenecks, and software inefficiencies. These problems become even more severe as data volumes continue to rise.
Recent advances in approximate computing have created new waves in solving both cost and speed problems. In many applications, perfect decisions can often be made without having perfectly accurate answers. Examples include feature engineering and selection, model tuning, visualization, and A/B testing, whereby 99.9% accuracy leads to the same decision as an exact answer, but can be obtained 200x faster and requires 200x fewer computational resources. In other applications, a partial but immediate answer can help determine whether a full computation is still needed, and thus dramatically reduce the computational footprint of an organization.
Despite recent strides in academia, there is little awareness about approximate computing in the industry. The ACAIA (Approximate Computing for Affordable and Interactive Analytics) workshop will provide a unique opportunity for valuable discussion and synergy on 1) what it means to incorporate approximation in existing data analytics frameworks, 2) how to cater to a wide range of users with different levels of statistical background, and 3) how to dramatically reduce the computational footprint of an organization.
Academic and industry influencers will give introductory presentations on fundamental concepts and techniques in approximate computing. Through a series of live demonstrations, participants will also receive hands-on training in approximation strategies using common technologies and popular frameworks. Following these, the focus will shift to the delegates themselves, who will partake in a roundtable discussion to share insight and experience, express their own use-cases, identify adoption concerns, explore potential solutions, and present conclusions in a collaborative spirit. A final set of outcomes will then be drawn ahead of further networking.
This exclusive, one-day workshop will offer a unique and valuable opportunity for industry leaders to learn the state-of-the-art in approximate computing, actively participate in an intimate setting, and truly contribute to driving the future success of the industry as a whole.