• Check out our 1st open tournament on reproducible SW/HW co-design of deep learning (speed, accuracy, costs) at ASPLOS'18!
• ADAPT'16 program is now available online!
• New CK-based workflow to crowdsource GCC/LLVM multi-objective optimization is now available! It supports our long-term effort to enable collaborative and reproducible program optimization and machine learning!
• Keynote by Ed Plowman (Director of Performance Analysis Strategy at ARM): "Benchmarks vs The Zombie Apocalypse: A comparison"
• Congratulations to Andrei Lascu, PhD student from Imperial College London, who have received dividiti award - 200 euros cash for the artifacts shared in the open Collective Knowledge format!

Computing systems are rapidly evolving into heterogeneous many-core machines. Developing such systems and optimizing real applications is becoming extremely complex, tedious and time consuming due to an unprecedented number of available design and optimization choices across the whole software and hardware stack. Using outdated, non-adaptive, non-systematic and ad-hoc design and optimization methodology with non-representative benchmarks and data sets often results in a highly inefficient software and hardware with a lack of performance portability and scalability, waste or expensive computing resources and energy, and slowing down time to market for new techniques and products.

ADAPT is an interdisciplinary workshop to discuss and demonstrate practical and reproducible techniques, methodology and tools that can help convert existing or future software and hardware into adaptive, scalable and self-tuning systems. Such systems should be able to automatically improve their characteristics (execution time, energy usage, size, accuracy, reliability, bandwidth, adaptation time and memory usage) depending on an application and its input, available resources, run-time state of the system, and user requirements.

ADAPT topics include but are not limited to machine learning based autotuning, representative benchmarking, real application self-tuning, automatic performance modeling, self-tuning compilers, automatic bug detection, run-time adaptation, automatic fault tolerance, dynamic hardware reconfiguration, predictive scheduling, new programming models, green data centers, adaptive embedded devices, reproducible experimentation, and optimization knowledge sharing. You can check out accepted papers from the past ADAPT workshops here.

Sponsors and supporters

-

DIVIDITI provided cash prize of 200 euros for the highest-ranked artifact shared in CK format