BISH Bash @ Mathworks

MathWorks, the maker of MATLAB and Simulink, will be hosting the next BISH Bash on Thursday, 29th. Oct. Join us for an evening of talks, networking and delicious bites. Please see below for the agenda and abstracts.
Please write a comment to let us know whether you will be attending in person or virtually, as we are planning a quality dinner for all in-person attendees. We look forward to seeing you there! (3975 Freedom Circle, Santa Clara, CA 95054, directions below). If you prefer to attend virtually, please send me an mail (ftiong@mathworks.com), and I will reply with the access link.

Agenda

• 6:00pm: Networking
• 6:30pm: Loudspeaker placement with AI - Francis Tiong, MathWorks
• 7:00pm: Wavelet Scattering -- Wayne King, MathWorks
• 7:25 - 7:35pm: break
• 7:35pm: Deploying Neural network onto Qualcomm processors with Hexagon NPU - Qualcomm
• 8:00pm: Agentic AI with Matlab - Mihir Kavishwar, MathWorks
• 8:25pm: Room acoustics evaluations - Francis Tiong, MathWorks
• 8:50pm: Networking

Loudspeaker placement with AI - Francis Tiong, MathWorks
This lecture shows how modern AI-assisted workflows can accelerate room acoustics analysis, simulation, and audio system design without compromising engineering rigor. It covers how to model room response above and below the Schroeder frequency with ray tracing and finite element methods, evaluate room modes and loudspeaker placement. Evaluation is based on the predicted Mean Opinion Score (MOS-LQO) from ViSQOL released by Google.

Wavelet Scattering: From Robust Features to Generative Models - Wayne King, MathWorks
The scattering transform creates stable, shift-invariant representations by cascading wavelet filters and modulus operations, acting as a shallow convolutional network with fixed, interpretable filters. After introducing its main variants (time, joint time–frequency, and image) for machine learning, this talk focuses on scattering spectra and wavelet phase harmonics. While standard wavelet coefficients across different scales are nearly uncorrelated due to separate frequency bands, these statistics realign scales to measure cross-scale correlation. They capture non-Gaussian structures—such as intermittency, bursts, and time asymmetry—in a compact feature vector that is often much smaller than standard scattering output with no loss in discriminative power. A few application examples will be demonstrated.

Audio Deployment on Qualcomm Hardware - Qualcomm
Qualcomm’s AudioReach simplifies AI-enabled audio development by abstracting complex heterogeneous hardware—including Hexagon DSP, CPU, and NPU—through a familiar graph-based model. Developers can design, deploy, tune, and profile advanced audio and AI pipelines without deep embedded-systems expertise.

Agentic AI with Matlab - Mihir Kavishwar, MathWorks
This presentation explores how agentic AI can accelerate engineering workflows in MATLAB and Simulink without compromising engineering rigor, verification, or human oversight. It introduces the MATLAB MCP Server, MATLAB Agentic Toolkit, and Simulink Agentic Toolkit, showing how trusted tools, reusable skills, and domain-aware workflows enable AI agents to create code, troubleshoot models, generate requirements and architectures, orchestrate simulations, and improve maintainability. Through practical examples, attendees will see how MathWorks tools make external AI agents more reliable, privacy-conscious, and useful for real engineering work.

Room acoustics evaluations - Francis Tiong, MathWorks
The session also demonstrates how to assess acoustic performance with objective and psychoacoustic metrics including RT60, Speech Interference Level (SIL), Speech Intelligibility Index (SII), Speech Transmission Index (STI) and short-time objective intelligibility (STOI). We will see how these metrics are applicable in qualifying the acoustics of a room.