Insights from NeurIPS 2024
Reflections on the future of AI from NeurIPS 2024, covering inference-time compute, unconditional generation, hardware efficiency breakthroughs, and the evolving relationship between AI and creativity.
Writing on AI and our Future
Some of my long-form thoughts.
Mapping the Landscape of Deep Learning: Topic Discovery from Book Embeddings
I created a dataset of deep learning concept embeddings, extracted hierarchical topic structure, and created text and concept indexes across 158 books about deep learning comprising 50k book-pages.
An Intuitive Explanation of SGD vs Gradient Descent
An intuitive exploration of why Stochastic Gradient Descent often outperforms traditional gradient descent in machine learning optimization, from data efficiency, a focus on progress, and leveraging randomness.
Auditing Stable Diffusion with Perplexity
I used prompt engineering and RAG to have Perplexity's Default LLM design a process for a demographics fairness audit of the Stable Diffusion v2.1 text-to-image model.
Generative AI as an Ethical Theorist
I used prompt engineering and RAG to get LLMs to pose as ethical theorists and react to a contemporary AI ethics dilemma, the potential adoption of social robots within public facilities for vulnerable and special needs children.
Optical Pairing: Streaming K-Means for Zero-Shot IoT Communication
A patented computer vision and online machine learning algorithm enabling zero-shot optical communication from IoT devices to arbitrary mobile phones, robust to challenging outdoor lighting conditions and distances up to 25ft.
Breaking and Protecting Cryptographic Hardware: Power Analysis Side-Channel Attack and Mitigation
Implementing DES encryption on an FPGA, performing power analysis attacks to extract key information, and designing countermeasures. Graduate-level hardware security research as an undergraduate at UConn.
Predicting Radio Tower Structural Health from Cable Vibration Spectral Monitoring
Structural health monitoring system for cable-stayed radio towers using accelerometer data and DSP spectral analysis. Monitored tower cable response to wind loading, analyzed frequency spectra to track structural vibration modes and infer cable tension and component failures.