Bibliographic record
Abstract
Matrix multiplication is a fundamental building block of many high-performance computingapplications. One example is machine learning, where recent advancements have motivated the usage of ever larger matrices. Supporting this growth is difficult due to the exponential scaling of the compute requirements for matrix multiplication. Much effort has gone into lightening this burden, with sparsity being an area of particular focus. One approach for improving dense matrix multiplication through sparsity is theuse of “butterfly matrices”, which approximate a single dense matrix as the product of log(N) distinct sparse matrices with fixed sparsity patterns. Compared to dense matrix-matrix multiplication, butterfly matrix multiplication is advantageous as it scales with O(N^2 ∗ log(N)), which is a significant improvement over O(N^3). The fixed sparsity patterns of butterfly matrices allows their behaviour to be well studied, and their operation to be deeply optimised. While butterfly matrices offer many advantages, current implementations onGPUs are unable to take full advantage of them, as the sparsity patterns are at odds with conventional GPU I/O and cache designs. This thesis presents the acceleration of butterfly matrix multiplication on Field-Programmable Gate Arrays (FPGAs), using their reconfigurability to address the memory access problems. It covers the design and implementation of a dedicated butterfly accelerator, as well as the build flow used to deploy it. We create a testbed to evaluate the accelerator against a GPU performing dense matrix multiplication over matrix dimensions ranging from
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.011 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".