Maciej Szankin

Working at the intersection of deep neural network algorithms and hardware with the Intel Labs Intelligent Systems Research (ISR) group. Research activities currently include:

• Performing hardware-aware neural architecture search (NAS) algorithm research and development.
• Solving multi-objective optimization problems using evolutionary and genetic algorithms
• Designing methodology for hardware-aware mixed numerical precision search.
• Work with other business units to integrate AI research into their products.
• Demonstrate research efforts and deliver systems to other Intel business units outside of Intel Labs.
• Authored papers and filed patent applications on AI research efforts. Worked on
• MLPerf NAS-based submissions.

Skills & Technologies: Research and Development (R&D) · Deep Neural Networks (DNN) · Algorithm Optimization · Evolutionary Algorithms · PyTorch

• Co-organizer & Publicity Chair for 2018-2021 editions.
• Lead session and hands-on workshops on Deep Learning frameworks for 100+ audience.
• Developed classroom framework that enabled 200 participants to participate in hands-on workshops.
• Prepared technical challenges for participants.
• Administration and development of School's portal.
  

• Research and implementation of novel AI solutions with goal of driving Intel growth in smart home and low-power embedded space. This includes both ensuring DL/ML frameworks are taking full advantage of Intel hardware (CPU, VPU, & GPU), as well as impacting next generation products by driving development of software libraries.
• Published white-papers in IEEE conferences and journals. Worked on BERT's FP32/BF16 distributed training, which resulted in ACM/IEEE Supercomputing Conference publication and HPC-AI Technical Talk at Intel's HPC Pavilion.
• Built mathematical models for DNN models computational cycles estimation, which enabled early performance estimations.
• Enablement of Movidius Neural Compute Stick for Android. Helped guide firmware and driver development by providing requirements based on common use-cases and complex multimodel ML workflows for validation.
• Development of scheduler for AI workload distribution on Android-based device. Working prototype proved lower-power utilization while providing end user with better experience.
• Developed framework for Quantization-Aware Training and Post-Training Quantization for Intel NNP-I inference chip. Directly supported customer, proved successful working within tight deadlines and fast-paced atmosphere.

Skills & Technologies: Research and Development (R&D) · Deep Neural Networks (DNN) · Smart Home Automation · ML-oriented Android Mobile App Development · TensorFlow · Intel Movidius

Full-time (Since 1/2015):

• Worked upstream on improving scheduler stability and workload latency in OpenStack Nova.
• Contributed to cloud hardware inventory tracking subsystem, a key effort to increase OpenStack enterprise readiness.
• Lead effort to refactor and centralize configuration management system for OpenStack Nova.
• Elected by community for Nova Bug Team Coordinator position.

Intern (Since 4/2012):

• Worked on a scheduler, code profiling and optimizations for Intel Service Assurance Administrator, OpenStack tool for creation of software-defined infrastructure with enhanced service level objectives, which helped to achieve 70% speed-up.
• Worked on building custom Linux distribution with Real-Time kernel with Yocto project. Testing and evaluation of target metrics on a prototype x86-based microarchitecture emulated on FPGA;
• Test risk and execution management tool with ML-supported decision making. Based on historical data the tool allowed to schedule minimum amount of tests while maintaining highest overall validation confidence level.

Projects: OpenStack Nova; distributed metric collection system for storage clusters and clouds;

Projects (Intern): Intel Service Assurance Administrator; OpenStack; Test risk and execution management tool with ML;

Skills & Technologies: Open Source development · Clouds · Virtualization & Containers · Distributed Databases · RTOS · Machine Learning

• Worked with customers to understand needs and provide excellent service