Seminars
A Universal Operator for Continuous Mathematics: Hardware Based on EML
Description
A NAND gate alone is sufficient to implement any Boolean function. Recent work proposes an analogous primitive for continuous mathematics: the binary operator EML(x,y) = exp(x) - ln(y), together with the constant 1, can generate arithmetic operations, exponentiation, logarithms, trigonometric functions, square roots, and other elementary functions.
This seminar investigates EML from a computer architecture and hardware-design perspective. Could computational systems based around a single universal mathematical operator provide advantages through highly regular processing elements or instruction sets? Possible implementations in digital hardware, FPGAs, or analog computing should be investigated and compared with conventional arithmetic units, CORDIC, lookup tables, and function approximation techniques. Numerical accuracy, implementation cost, latency, energy efficiency, and potential application domains should be considered.
References:
- Odrzywo?ek, “All Elementary Functions from a Single Binary Operator,” 2026.
- Related literature on CORDIC architectures, elementary-function units, arithmetic accelerators, and function approximation hardware.
Supervisor:
Can Better Hardware Make Branch-and-Bound Faster?
Description
GPUs have transformed machine learning and scientific computing, yet many mixed-integer optimization problems remain strongly dependent on CPUs. Branch-and-bound exhibits irregular control flow, dynamically evolving search trees, synchronization, and complex memory-access patterns that are difficult to map efficiently onto conventional massively parallel architectures.
This seminar investigates how modern and emerging computing architectures can accelerate branch-and-bound and related optimization algorithms. Relevant architectures include many-core CPUs, GPUs, heterogeneous CPU-GPU systems, distributed systems, FPGAs, and specialized accelerators. You should identify the dominant architectural bottlenecks of branch-and-bound and investigate which solver components benefit from parallel or specialized hardware. Particular attention should be given to the interaction between algorithms, memory systems, parallelism, and workload characteristics.
References:
- Helbecque et al., “Portable PGAS-Based GPU-Accelerated Branch-and-Bound Algorithms at Scale,” Concurrency and Computation: Practice and Experience, 2025.
- Tezoto et al., “Performance and Portability in Multi-GPU Branch-and-Bound: Chapel Versus CUDA and HIP for Tree-Based Optimization,” IPDPSW 2025.
- Tardivo, Michel, and van Hoeve, “GPU-Accelerated Relaxed Decision Diagrams for Branch-and-Bound Optimization,” CP 2026.
- NVIDIA, “Learn How NVIDIA cuOpt Accelerates Mixed Integer Optimization Using Primal Heuristics,” 2026.
- Recent literature on parallel branch-and-bound, heterogeneous optimization, and hardware acceleration of combinatorial optimization.