Last summer\u2019s paper, by Duan, Zhou and Wu, showed that Strassen\u2019s process could still be sped up significantly. It was all because of a concept they called a hidden loss, buried deep within previous analyses \u2014 \u201ca result of unintentionally killing too many blocks,\u201d Zhou said.<\/p>\n

The laser method works by labeling blocks with overlaps as garbage, slated for disposal; other blocks are deemed worthy and will be saved. The selection process, however, is somewhat randomized. A block rated as garbage may, in fact, turn out to be useful after all. This wasn\u2019t a total surprise, but by examining many of these random choices, Duan\u2019s team determined that the laser method was systematically undervaluing blocks: More blocks should be saved and fewer thrown out. And, as is usually the case, less waste translates into greater efficiency.<\/p>\n

\u201cBeing able to keep more blocks without overlap thus leads to \u2026 a faster matrix multiplication algorithm,\u201d Le Gall said.<\/p>\n