Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
INESC-ID, Instituto Superior Técnico Universidade de Lisboa,Portugal
3
3
5
3
PIM-based acceleration can deliver 2.6x speedups for sparse tensor decomposition, but distributed memory limitations in current PIM architectures can still be a bottleneck.
Constant-depth threshold circuits can unlock efficient, pipelined epistasis detection on neuromorphic hardware, sidestepping the complexity bottlenecks of classical approaches.
Finally, a single tool automatically generates accurate Cache-Aware Roofline Models across Intel, AMD, ARM, and RISC-V CPUs, enabling portable performance analysis.