Issue link: https://iconnect007.uberflip.com/i/1532533
FEBRUARY 2025 I PCB007 MAGAZINE 9 Marcy LaRont is the editor of PCB007 Magazine. Marcy started her career in PCBs in 1993 and brings a wide array of business experience and perspective to I-Connect007. To contact Marcy, click here. On the program side, Udo Welzel and Stan- ton (Stan) Rak, chairs of the Technical Program Committee, share insights into this year's Techni- cal Conference, as well as information on the two special sessions on advanced packaging. Teresa Rowe, IPC senior director of assembly and stan- dards technology, talks about why standards committee work is so important and rewarding. omas Marktscheffel discusses standards work specifically for Factory of the Future, explaining the various elements from a standards perspec- tive. Kelly Scanlon, IPC's lead sustainability strat- egist, gives a thorough year-in-review on sustain- ability, and previews what's next for IPC in this important area. Finally, we highlight the IPC Emerging Engineer program with a spotlight on Hannah Grace and Paige Fiet, two alumni of the program now working in the industry. Dan Feinberg continues his special series on IPC Hall of Famers, this month acknowl- edging Denny Fritz. PCBAA chair Shane Whiteside continues his discussion on how our business sector must deal with the cur- rent environment in Washington, D.C., stay- ing laser-focused on our goals for U.S. PCB fabricators. And we have some fun infograph- ics this month looking back at 25 years, which I know you will enjoy. When you say it out loud, 25 years sounds like a long time, but in many ways feels like the blink of an eye. Yet, what has been accom- plished, the arc of technology over these past two and a half decades, has had an impact that is nothing less than epic. Look for these trends to continue. So, let's begin to "Reimagine the Possibilities" and learn more about IPC APEX EXPO 2025: What you have to look forward to, and what you don't want to miss. PCB007 Artificial intelligence (AI) is everywhere, from the chatbots we consult for customer support to tools predicting how diseases might spread. But the computing power and energy required to power modern AI models—such as large lan- guage models (LLMs)—can make them expen- sive, inaccessible, and environmentally tax- ing. A team of researchers at Rice University is working on solutions to change that. "Generative artificial intelligence is still in its infancy when it comes to broader integration," said Anshumali Shrivastava, associate profes- sor of computer science, electrical and com- puter engineering and statistics and member of Rice's Ken Kennedy Institute. Shrivastava and several members of his re- search group presented three of their most re- cent advancements in tweaking LLMs to bet- ter suit users' needs at the latest convening of the flagship AI conference Neural Information Processing Systems, or NeurIPS, in Vancouver, British Columbia, last December. The three pa- pers develop superior alternatives to popular strategies such as low rank approximations and standard quantizations, showcasing the impact potential and creativity of AI research at Rice. One of the Rice team's papers presented at NeurIPS explores a concept Shrivastava calls "parameter sharing," introducing Sketch Struc- tured Transforms (SS1) — a method for handling the vast tables of numbers, called weight ma- trices or working memory, that AI models rely on to make predictions and decisions. For ex- ample, when applied to popular LLMs, the SS1 technique sped up processing times by over 11% without requiring additional fine-tuning. Shrivastava's team has developed an algorithm that allows LLMs to run efficiently on standard computer processors (CPUs) instead of GPUs. In a third paper, the team introduced "cou- pled quantization," a method for compressing this memory without losing the quality of the model's responses. Shrivastava's work reflects a broader vision for the future of AI, one where advanced AI is available to everyone, not just tech giants. Source: Rice University Rice Computer Scientists Develop Solutions for Making AI Models More Efficient and Customizable