Hi! I'm Jeffrey Mei, an applied scientist at SiriusXM/Pandora.

My current work is in using transformers and self-attention to develop better recommender models for music personalization. Previously, I worked on incorporating style into next-item recommendations at Wayfair, and my PhD research focused on investigating sea ice deformation using computer vision and deep learning techniques with remote-sensing data. The unifying theme of both my academic and industry research is in improving interpretability of "black-box" deep learning models, along with a passion for elegant data visualization.

Portfolio

My doctoral thesis research with my PhD supervisor, Dr. Ted Maksym (WHOI) involved quantifying sea ice thickness, which cannot be easily measured by satellites, using lidar-measured surface elevation. I improved current statistical methods of inferring sea ice thickness by using deep learning techniques, such as convolutional neural networks, to distinguish different surface topographies (example) and to relate the surface texture to the local sea ice thickness. I also worked with computer vision techniques, such as delineating sea ice floes in satellite imagery using OpenCV (example), or classifying sea ice topographies based on their textures.

In 2017, I took part in the PIPERS expedition to the Ross Sea during Antarctic winter to collect snow depth and sea ice thickness data (example). This work was ultimately published in a journal article.

In the past, I worked on localizing glacial calving by using numerical methods and signal processing with local seismic data. In 2015, I conducted field work in Sermilik Fjord, Greenland to collect oceanographic and seismic data near Helheim Glacier, which was eventually published.

Random side projects I have done include simulating the sea ice ridging process using an open-source particle simulator (LIGGGHTS); I have also been compiling various scripts useful for handling cryoscience data that I have developed during my PhD into a set of open-source tools (cryo-toolbox).

I spent two years at Wayfair as a ML scientist in Product Recommendations. See here for an example of my work! I am now at SiriusXM/Pandora working on music recommendations.

(Accepted to RecSys 2022) An interactive visualization of item embeddings learned by a next-item recommender model that I developed at Wayfair. The training data consists simply of sequences of browsed items for a large number of customers; the model learns to group different types of furniture that share certain attributes. These include style ("Victorian"), material ("glam" - which is largely premium materials like glass, marble and velvet; "jute") and functionality ("no floor space" - including storable beds, convertible sofas and wall-mounted TV stands).

Training an autoencoder to reproduce an input surface elevation (laser altimetry) scan using PyTorch. An unsupervised clustering technique can then be applied to the learned feature vectors to infer the different classes (types) of surfaces.

A "layer cake" of snow depth, sea ice freeboard and sea ice draft (top to bottom) to develop a geospatial model of snow distribution along Antarctic pressure ridges. Visualized with Axes3D and Matplotlib and animated with ffmpeg.

Interactive sea ice floe detector, using sea ice imagery from NSIDC. Click on image to view higher-quality animation. Built using OpenCV and Python.

Developing discrete-element models for pressure ridge formation using LIGGGHTS. Colorbar shows stress (kPa).

Publications

Peer-reviewed Papers

- Mei, M. Jeffrey, Oliver Bembom, Andreas Ehmann. "Station and Track Attribute-Aware Music Personalization" In Seventeenth ACM Conference on Recommender Systems (RecSys ’23) 1031-1035, 10.1145/3604915.3610239, 2023.
- Mei, M. Jeffrey, Cole Zuber, Yasaman Khazaeni. "A Lightweight Transformer for Next-Item Product Recommendation" In Sixteenth ACM Conference on Recommender Systems (RecSys ’22) 546-549 10.1145/3523227.3547491, 2022.
- Mei, M. Jeffrey, Ted Maksym. "A Textural Approach to Improving Snow Depth Estimates in the Weddell Sea." Remote Sensing 12, 1494, 10.3390/rs12091494, 2020.
- Mei, M. Jeffrey, Ted Maksym, Blake Weissling, Hanumant Singh. "Estimating Early-Winter Antarctic Sea Ice Thickness From Deformed Ice Morphology." The Cryosphere, 13, 2915–2934, 10.5194/tc-13-2915-2019, 2019.
- Mei, M. Jeffrey, David M. Holland, Sridhar Anandakrishnan, Tiantian Zheng. "Calving localization at Helheim Glacier using multiple local seismic stations." The Cryosphere, 11, 609-618, 10.5194/tc-11-609-2017, 2017.
- Holland, D. M. , D. Voytenko, K. Christianson, T. H. Dixon, M. J. Mei, B. R. Parizek, I. Vankova, R. T. Walker, J. I. Walter, K. Nicholls, and D. Holland. "An intensive observation of calving at Helheim Glacier, East Greenland." Oceanography 29(4):46–61. 10.5670/oceanog.2016.98, 2016.

Selected Talks

- Industry talk, Sixteenth ACM Conference on Recommender Systems , 2022
- Keynote talk, Context-Aware Recommender Systems Workshop, 2022

Theses

- Doctoral Dissertation: 'Morphological Approaches To Understanding Antarctic Sea Ice Thickness', 2020.
- Undergraduate Capstone: 'A Novel Two-Station Seismic Method to Locate Glacier Calving', 2015.

Selected Conference Publications

- Mei, M. Jeffrey, Ted Maksym, “A Textural Approach to Snow Depth Distribution on Antarctic Sea Ice.” European Geophysical Union General Assembly. Online. 2020. [link]
- Mei, M. Jeffrey, Ted Maksym, “Estimating Early-Winter Antarctic Sea Ice Thickness From Deformed Surface Morphology.” International Glaciological Society Sea Ice Symposium. Winnipeg, Canada. 2019. [link]
- Mei, M. Jeffrey, Ted Maksym, Arnold Song, Matthew Parno, Guy Williams, Hanumant Singh, Jeffrey Anderson, Alek Razdan. "PIPERS: Sea Ice Thickness Redistribution From Early Winter Deformation." 2018 SCAR/IASC Conference. Davos, Switzerland. 2018. [link]
- Mei, M. Jeffrey, Tiantian Zheng, David M. Holland. "A Novel Seismic Method for Glacial Calving Localization." American Geophysical Union Fall Meeting 2015. San Francisco, USA. 2015. [link]
- Mei, M.Y. Jeffrey, I. Zaw, and L. J. Greenhill. "Correlations of Circumnuclear Water Maser Luminosity with AGN Activity and SMBH Mass." American Astronomical Society Meeting Abstracts #223. Vol. 223. Washington D.C., USA. 2014. [link]

Other Projects

- HiggsHunters, a Zooniverse citizen science project developed by the University of Oxford and New York University to analyze particle physics data from the ATLAS experiment at the Large Hadron Collider at CERN. Coverage of the launch in November 2014 in WIRED and CERN.
- Cryo-Toolbox, an open-source set of tools for other sea ice researchers, especially those using data from the National Snow and Ice Data Center.

About Me

Deploying moorings in Sermilik Fjord, Greenland in August 2014.

Taking sea ice thickness measurements in the Ross Sea during PIPERS 2017.

I attended Westlake Boys High School in Auckland, New Zealand, where I completed Cambridge International Examinations A levels in Mathematics, Physics, Chemistry and German Literature and also received an Outstanding Scholar award from the New Zealand Qualifications Authority, with Scholarships in English, Art History, Chemistry and Mathematics with Calculus, and an Outstanding Scholarship in Statistics and Modelling.

For my undergraduate education, I was offered a full scholarship to New York University Abu Dhabi, a small liberal arts college in the United Arab Emirates, where I graduated with a Bachelor of Science cum laude with a double major in Mathematics and Physics. During this time I spent two semesters studying abroad, one at each of NYU Berlin/Humboldt Universität zu Berlin and NYU Washington Square. I have done summer research with Dr. Ingyin Zaw at the Center for Cosmology and Particle Physics at NYU, Dr. David Hogg at the Max-Planck Institut für Astronomie in Heidelberg, Dr. Andy Haas at the Center for Cosmology and Particle Physics at NYU and Dr. David Holland at the Courant Institute of Mathematical Sciences at NYU.

After that, I completed a PhD in Oceanographic Engineering at the MIT/WHOI Joint Program in Applied Ocean Science & Engineering working under Dr. Ted Maksym (WHOI). My research involved applying deep learning and computer vision techniques to solve sea ice problems. In my spare time, I also served as the President for the MIT Badminton Club and the Kiwis@MIT student group.

I spent two years as a ML Scientist at Wayfair, working on transformer-based next-item recommendations (see Portfolio). As of now, I am working at SiriusXM/Pandora on music recommendations.

Contact

The best way to get hold of me is via email, (jeffrey [dot] mei [at] alum [dot] mit [dot] edu).