google deutsche bank cloud deal reuters 1594109955741

Google fires another AI researcher after an internal battle over chip design research

Google on Monday fired a senior engineering manager after allegations that co-workers who were doing milestone research on artificial intelligence software were trying to humiliate him and harass his behavior.

The controversy, which stems from attempts to automate chip design, risks undermining Google’s reputation in the academic community. This could disrupt the flow of millions of dollars in government grants for research on AI and chips.

The Google Research Unit has been under scrutiny since the end of 2020 after workers made public criticisms about the handling of staff complaints and publishing practices.

The new episode follows the publication in June of the scientific journal Nature of “Graph Placement Methodology for Faster Chip Design” led by Google scientists Azalia Mirhozeni and Anna Goldie. They found that AI could complete a key step in the chip design process known as floorplanning, and that the unspecified anthropologist would be faster and better than the subjective reference point.

But in a paper posted anonymously by other Google colleagues online anonymously in March – “Strong Baselines for Evaluating Deep Reinforcement Learning in Chip Placement” – two alternative approaches based on basic software found better performance than AI. One beat in the familiar test, the other in the proprietary Google rubric.

Google declined to comment on the leaked draft, but two workers confirmed its authenticity.

The company said it refused to publish Stranger Baseline because it did not meet its standards, and immediately fired Satrajit Chatterjee, a prominent driver at work. Refused to say why he was fired.

Chatterjee’s lawyer Larry Burgess said: “It is unfortunate that Google is taking this turn. “His goal is to be transparent about science, and he asked Google to address this within two years.”

Google researcher Goldie told the New York Times that the shooting was the first of its kind on Monday, and that Chatterjee had been harassing her for years by spreading false information about her and Mirhose.

Burgess denied the allegations and added that Chatterjee had not leaked strong baselines.

Patrick Madden, an associate professor who focuses on chip design at the University of Binghamton after reading both papers, says he has never seen a paper that does not have a good resemblance to Nature.

“It’s like a reference problem: everyone gets the same jigsaw puzzle pieces and you can compare how close you got to getting everything right,” he said. “If they provide results on some standard benchmark and they are stars, I will sing their praises.”

Google said the comparison with the human was very relevant and that software licensing issues prevented the tests from being specified.

Studies by large organizations such as Google in popular journals have a profound effect on whether or not they finance similar projects in the industry. A leaked paper by a Google researcher has unfairly opened the door to questions about the credibility of any work published by the company.

Following the emergence of “Stronger Baselines” online, Jobin Ghrahmani, Vice President, Google Research, wrote on Twitter last month, “Google stands for this work published in Nature in ML for Chip Design, which is a replica of and used by Google on OpenSource.”

Nature did not immediately comment, citing a UK public holiday. Madden said he hopes Nature will revisit the publication, noting that the peer reviewer notes asked for at least one result on the benchmarks.

“Something, it never happened,” he said.

© Thomson Reuters 2022


Leave a Comment

Your email address will not be published.