October Section Meeting/Joint UW-Madison Student Branch Meeting: (Photos of the event are available on the links by Tom Kaminski and Charles Cowie.) The talk was preceded by a display of some high-end vehicles equipped with autonomous driving systems. The vehicles were on the walk between Lot 17 and Engineering Hall on a path with a lot of student foot traffic, generating an additional draw to the meeting. The meeting was well attended and Bob Neff gave a presentation that covered the history of autonomous vehicles up to the present. Bob also discussed several of the challenges facing adoption of the technology. Despite the US Government Department of Transportation’s early standards, auto manufacturers were slow to adopt to the technology in a “chicken and egg” struggle. Who would build a car to drive on a modern highway if there were no highways? Advances in technology have now made it possible to drive autonomously on modern roads, but as of today, no country in the world allows a driverless auto to operate on public highways.
October Special Meeting: Dr. James Smith, Emeritus Professor of Electrical and Computer Engineering, gave a talk to the College of Engineering Students on the topic “Space-Time Computing and the Brain”. Dr. Smith based his talk on the advances in the neuroscience community that have been describing how the brain operates and claimed that Computer Architects should be shamed by the advances in the neuroscience community that have been ignored by the engineering community. He challenged the Computer Architects in the audience to quit “chasing the bright, shiny object” of deep learning neural networks and get to work on the more significant problem of finding a way to mimic the way the brain really computes.
Dr. Smith showed a taxonomy of the current neural network models and introduced the time-space model. He then went on to motivate his discoveries by showing a simple combinatorial network that transforms a vector of zero’s (binary digits) to a vector of zero’s, yet sorts the arrival times of the 1 to 0 transitions in sequence. He then proposed a mathematical algebra based on “Space-Time” computing and showed how it closely mimics true neural networks. Dr. Smith used introduced space-time algebraic primitives to build neural networks that behaved in the manner as all of the current models based on neuroscience research. Finally, he proposed that these primitives could be built from existing CMOS on silicon technology without the need to invent new device technology.
November Life Member Affiliate Group Meeting: Steve Schultheis gave a detailed and informative talk on the history of musical instruments and the evolution of electronic instruments, some of which attempted to mimic the traditional instruments and some of which created unique sounds of their own. Steve spent a good amount of time discussing the details of a simple microphone, showing how that device evolution over time did not substantially improve on the earliest devices. He discussed a number of trade offs in the design of the electronic instruments such as the “sampled” (huge files required) verses “simulated” (huge computational requirements) approaches to instruments. While the sampling technique works well for percussion instruments, such as drums, pianos require huge files sizes, on the order of a half-terabyte. Sampling has little hope of duplicating the interaction of multiple keys pressed at once. During the presentation, Steve used his programmable equipment to show how different sounds could be synthesized. Steve finished the presentation with a short concert, playing a jazz tune. A printed copy of Steve’s Presentation is available here.