Concept: Latvia-based startup has unveiled the AutoDL framework, an AI-driven optimization technology for neural network compression and acceleration. It supports deep neural network optimization for AI developers and edge AI organizations.

Nature of Disruption: The AutoDL framework includes Python API that can be integrated into many neural network training workflows quickly and easily. It automates the search for the best network design by considering several hardware and software characteristics including latency, RAM (random-access memory), model size constraints, and operation type. framework takes a trained neural network as input and determines the shortest subnetwork with the lowest latency and without accuracy deterioration. This technique makes the neural network shorter, faster, and more energy-efficient, addressing concerns that developers face. The startup uses all of these strategies at the same time to obtain the highest compression/acceleration rate while maintaining accuracy. According to the startup, its technology can enable AI developers and businesses to achieve up to 20 times neural network acceleration and 25 times network compression. The advantages can potentially help reduce computing hardware costs by up to 70%.

Outlook: Neural networks are currently frequently employed in production and applications. However, it must be more efficient in terms of computational resource consumption and more economical. Developers are looking for solutions that can make its implementation faster, better, and less expensive.’s AI optimization solution addresses this issue, allowing for rapid, real-time AI advancement. This technology aids customers in achieving considerable cost reductions and much speedier product launches, as well as shortening time to market. The industries that deploy’s framework include healthcare, IoT, electronics, oil & gas, mobile apps, autonomous driving, telecom, cloud computing, and robotics. It claims to have worked on pilot projects with over 20 firms, including Sony, PicsArt, Huawei, LG, Dscribe, and Hive, among others.

This article was originally published in