Albert Haque
I'm a PhD student at Stanford studying artificial intelligence, advised by Fei-Fei Li. My technical interests combine machine hearing, computer vision, and natural language processing to improve the physical spaces of healthcare. You can read about my work in The Wall Street Journal and MIT Technology Review. In industry, I've worked on Tesla Autopilot and ads teams at Google, Facebook, and Microsoft. I did my BBA in finance and BS in computer science at the University of Texas at Austin. I love traveling and fly around 100K miles to 10 countries each year.
Research
Spoken Sentence Embeddings
Saying the same word with different tones can change its meaning. It is difficult for text alone to capture this. We created audio-linguisitc embeddings for this purpose.
ICASSP 2019
Albert Haque, Michelle Guo, Prateek Verma, Li Fei-Fei
Detecting Depression from Speech and Faces
Cost and social stigmas prevent some people from receiving mental care. With the goal of deploying to phones, we detected depression using voice and 3D facial expressions.
NeurIPS 2018 Workshop on Machine Learning for Health
Albert Haque, Michelle Guo, Adam Miner, Li Fei-Fei
Automatic Medical Transcription
Medical diagnosis codes are hard to pronounce. We trained a RNN to automatically transcribe medical ICD codes when the microphone is far away.
NeurIPS 2018 Workshop on Machine Learning for Health
Albert Haque, Corinna Fukushima
Activity Recognition from Low Resolution Images
One way of preserving privacy is by using low-resolution images. We trained an algorithm to identify clinical activities from 14 x 14 pixel images using super-resolution.
NeurIPS 2018 Workshop on Machine Learning for Health (Spotlight)
Edward Chou, Matthew Tan, Cherry Zou, Michelle Guo, Albert Haque, Arnold Milstein, Li Fei-Fei
Voice and Musical Style Transfer
Our model can take any word or musical note and transform it into any target voice or instrument, as long as it has heard the voice or instrument before.
Interspeech 2018 (Oral)
Albert Haque*, Michelle Guo*, Prateek Verma
Inference on Encrypted Images
Homomorphic encryption (HE) allows algorithms to make predictions from encrypted data. We use HE to detect diabetic retinopathy from encrypted retinal photographs.
ICML 2018 Workshop on Privacy in ML and AI
Josh Beal*, Edward Chou*, Albert Haque, Li Fei-Fei
Dynamic Task Priority for Multitask Learning
In computer vision, is classification the same difficulty as pose estimation? What about segmentation? In this work, we dynamically prioritized different tasks during training.
ECCV 2018
Michelle Guo, Albert Haque, De-An Huang, Serena Yeung, Li Fei-Fei
Identifying Risky Hand Hygiene Scenarios
Risky hand hygiene scenarios happen when the patient may be exposed to new bacteria. We counted and predicted when these scenarios happen.
NIPS 2017 Workshop on Machine Learning for Health
Michelle Guo, Albert Haque, Jeff Jopling, Lance Downing, Alexandre Alahi, Brandi Campbell,
Kayla Deru, William Beninati, Arnold Milstein, Li Fei-Fei.
Measuring ICU Patient Mobility
Intensive care units (ICUs) house patients with life-threatening conditions. In this work, we measure how often patients move around. This can monitor the recovery process.
NIPS 2017 Workshop on Machine Learning for Health (Spotlight)
Gabriel Bianconi, Rishab Mehra, Serena Yeung, Francesca Salipur, Jeff Jopling, Lance Downing, Albert Haque,
Alex Alahi, Brandi Campbell, Kayla Deru, Bill Beninati, Arnold Milstein, Li Fei-Fei
Detecting Hospital Hand Hygiene
Hand hygiene is very important for hospitals and has been linked to hospital-acquired infections. We created a 3D computer vision algorithm that beats in-person auditors.
MLHC 2017 (Spotlight)
Albert Haque, Michelle Guo, Alex Alahi, Serena Yeung, Zelun Luo, Alisha Rege, Amit Singh, Jeff Jopling, Lance Downing, Bill Beninati, Terry Platchek, Arnold Milstein, Li Fei-Fei
Pose Estimation from Top-Down Cameras
Pose estimation identifies the position of a person's body parts. Most algorithms work from clean, side views. We made an algorithm that works from top-view cameras.
ECCV 2016
Albert Haque, Boya Peng*, Zelun Luo*, Alexandre Alahi, Serena Yeung, Li Fei-Fei
Detecting Trolls on Twitch Plays Pokemon
Thousands of internet users simultaneously entered commands to play a single game of Pokemon in real-time. We tried to detect who was trolling or not.
Technical Report, University of Texas, 2014
Albert Haque
Distributed Graph Databases
Traditional databases are optimized for tabular data. We created NoSQL graph databases and analyzed data serialization, internal structure, and query speed tradeoffs.
ISWC 2013
Philippe Cudre-Mauroux, Iliya Enchev, Sever Fundatureanu, Paul Groth, Albert Haque, Andreas Harth, Felix Leif Keppmann, Daniel Miranker, Juan Sequeda, Marcin Wylot