#research
high-performance computing for ML and scientific workloads — across GPUs, SIMD, FPGAs, and RISC-V.
hpc for drug discovery on intel max gpus
Optimizing the LiGen virtual-screening pipeline for Intel Max (Ponte Vecchio) GPUs. The goal is to close the portability gap — SYCL code that runs well on Intel without giving up the performance we'd get from vendor-native paths.
with B. Cosenza, L. Carpentieri, A. De Caro · PDP 2026 · collaborators at Politecnico di Milano & Dompé.
quantized inference on AVX & RVV
A careful, architecture-level look at power-of-two quantization for neural-network inference on wide-SIMD CPUs. We benchmark across AVX (x86) and RVV (RISC-V) to understand which gains come from the quantization scheme itself, and which from the instruction set.
with G. Pagano, B. Cosenza · ITADATA 2025 Workshops.
fpga acceleration of 3d deep learning
Accelerating point-cloud classifiers (DGCNN) on FPGAs using HLS and Verilog. The underlying question: how far can we push irregular, graph-structured DL models onto reconfigurable fabric without giving up accuracy?
MSc thesis · published in Circuits, Systems, and Signal Processing (Springer, 2023).
# if you're a prospective collaborator, the fastest way in is email with a concrete question.