Maxwell Mckinnon Email & Phone Number

Mountain View, Ca, Us

Cupertino, California, Us

Audio, Machine Learning, On-Device Deep Learning (Distillation, Quantization, etc.), Cloud-Backed Deep Learning, Data Science, Analytics, NLP

San Jose, Ca, Us

Standout Projects:♦ Designed production schematic and production layout for a low power (~1.5W) class D audio amplifier targeted for wearables. Led four-person evaluation team♦ Wrote a testing manual and trained five engineers from our China team on testing and supporting one of our best selling parts. This enabled our China team to better support Chinese customers and saved me time from repeatedly debugging the same types of problems.♦ Built and presented demos at CES 2017 in Chinese and English: Dynamic Speaker Management (DSM) and Imperceptible Power Reduction (PPR). DSM is an algorithm for microspeakers found in watches to laptops that simultaneously maximizes loudness and audio quality (THDN & Rub/buzz) while protecting the speaker from heat failure and excursion failure, the two dominant failure modes for speakers of this size. PPR is similar in concept to wav->mp3 compression but instead of imperceptible information reduction (or perceptible depending on the bit rate~), it's optimized for imperceptible power reduction (not just a simple high pass filter, although that's part of it) on a battery powered device.♦ Developed two significantly more precise automation tools by using a different methodology on the same equipment we already had. Both are now standard routines used by others in the lab: 1. Measuring above 80dB PSRR now up to 100dB PSRR (note: PSRR and crosstalk are the same thing) 2. Tuning the measurement of INL/DNL of an ADC faster and more precisely, a change from waiting a day to waiting an hour. This was very much like a software optimization problem with noise, local optimums, etc. - each tuning step took 20 minutes and the steps were linear (just guessing / hill climbing algorithm). Now each step takes 5-20 minutes depending on the necessary precision needed at that point in the optimization. The optimization algorithm also now uses gradient descent (much faster!) with annealing to get out of local optima from the measurement noise.

San Jose, Ca, Us

Standout Projects:♦ Designed and developed the first resuable code GUI where any GUI could be generated on day one of receiving silicon, never done before in our group. Additionally, the main data structures and project architecture were reused in several key projects: a Linux device driver generation tool and a generic bench automation tool♦ Wrote a testing manual and trained five engineers from our China team on testing and supporting one of our best selling parts. This enabled our China team to better support Chinese customers and saved me time from repeatedly debugging the same types of problems.Routine Tasks:♦ Automate test equipment (AP, APx, C#, python, VBA) and work closely with customers and engineers from design, drivers, and test to evaluate and debug audio ICs♦ Support English and Chinese speaking customers and facilitate debug both over email, calls, and onsite visits.

Moscow, Id, Us

♦Worked in the area of bioinformatics.♦Built a tool in python to model the degree of epistasis in a biological system using Walsh transforms.♦Built another tool in python to compute the automorphism classes of N-dimensional epistasis, up to 4-dimensional space. Proved mathematically that 5D required between 10^17 and 10^20 bytes to store the automorphism class list (uncompressed). The breakthrough from brute forcing 4D (projected time many years) to a solution that finished in a few hours came from transforming the main question the algorithm answers from "yes-no" to "maybe-no", an incredibly simple change. "yes-no" is a very costly question to ask, and it didn't need to be asked for the majority of cases. The expensive "yes-no" operation could then be performed on the remaining small list of maybes. This simple yet powerful epiphany was an awesome moment!It was unproven if the algorithm was correct however, so there was more work to be done. I followed up and proved the 4-dimensional solution was correct in size mathematically through chromatic polynomials and Burnside Lemma made manageable with Matlab scripts and Mupad algebraic tools.

San Jose, Ca, Us

♦ Built and programmed a lazy Susan for audio testing from the ground up with laser-cut acrylic, robotic components and drivers, Arduino, and Processing♦ Tested Audio ICs with Audio Precision♦ Implemented a digital two microphone and two source noise separation algorithm based upon ICA.

Moscow, Id, Us

♦I taught the C++ lab for first year students as well as held tutoring hours in the lab. ♦This class wasn't about teaching C++ as much as teaching thinking with computers. To me, my most important job was to convince the students who had never seen programming before that they were doing fine and only needed to work hard and have fun. The attrition rate for beginning computer science classes is absurd due to the widespread skill level of incoming students as well as disconnect from learned material to potential application. My goal was to teach and encourage the true beginners, challenge the experienced, and cultivate self motivation and interest in everyone through examples of interesting applications.

Klamath Falls, Oregon, Us

♦ Promoted the well being of residents and encouraged student leadership in the residence halls. Served as a positive role model and leader for students. ♦ Catered to the special needs of residents and made referrals to appropriate campus professionals. ♦ Raised the smile to frown ratio to record highs thanks to groundbreaking BBQ breakfast techniques.

Klamath Falls, Oregon, Us

♦ Fostered academic success♦ Lead workshops for improving specific skills♦ Maintained flexibility for both drop-in and appointment tutoring.

Master Of Science, Computer Science

Bachelor Of Science X2, Electrical Engineering, Applied Mathematics