MLconf Online 2020
2020-11-05..2020-11-06
-- online --
| title | author | length |
|---|---|---|
| Heart to Heart and Eye to Eye with Deep Fakes | Ilke Demir | 32m |
Uses pulse-derived PPG and gaze geometry as authenticity signals to detect and attribute portrait deep fakes, not just pixel artifacts. Extracts remote PPG… | ||
| Challenges in Real-Life Face Recognition | Stanislaw Sielicki | 24m |
We study a Face Recognition problem caused by unforeseen in the training set variations in face images, related to artistic makeup and other occlusions… | ||
| LINC: AI for the conservation of the African lion | Agustin Mautone | 25m |
LINC builds a working pipeline: a Faster R-CNN object detector for faces and whisker regions, a ResNet-50 trained classifier turned 512-d embedding with k‑NN… | ||
| Scalable High Precision OCRs for Game Event Detections | Topojoy Biswas | 28m |
A production pipeline converts cheaply generated noisy labels into domain-specific models that locate broadcast clocks and read their text with near 100%… | ||
| Debiasing Contextualized Word Embeddings in NLP | Paul Barba | 20m |
Lexalytics and UMass rewrite a large corpus (replace Mister→prefix, sister→sibling, neutral pronouns) and retrain contextual models (ELMo) to reduce gender… | ||
| Designing More Inclusive AI with More Inclusive Workflows | Dan Gifford | 27m |
Finds that 90% of top "nurse" results are women and that this matches US census numbers, then questions whether matching a biased status quo is the right goal… | ||
| Mathematical Approaches to Clustering | Joseph Ross | 26m |
Compares concrete algorithms and mathematical frameworks and gives clear takeaways: k-means fails on elongated strip data because it enforces convex Voronoi… | ||
| Play | Vidhya Murali | 27m |
Spotify voice personalization serves ~299 million users over 16 million tracks and 1.5 million podcast titles and treats voice as an audio to audio, zero GUI… | ||
| Voice Assistants in Healthcare | Galina Grunin | 19m |
A reference architecture that delivers identical member and provider information across web, mobile, call center and voice devices (Amazon Echo, Google Home)… | ||
| Prediction Model for Favorable COVID-19 Patient Outcomes | Yin Aphinyanaphongs | 26m |
Retrospective model trained on 3,345 Covid-positive adults and prospectively validated on 474 predicts a favorable 96-hour outcome (no ICU transfer… | ||
| The COVID Scenario Pipeline: High Stakes Data Science | Josh Wills | 28m |
A volunteer team spins up a reproducible pipeline that turns the Johns Hopkins SEIR code into docker images, profiles and speeds it to run in minutes, and then… | ||
| AI for COVID-19: An online virtual care approach | Xavier Amatriain | 29m |
Mobile-first chat app and custom EHR integrate AI tools for virtual care in California, with provider escalation built in. Uses an expert-system+deep-learning… | ||
| Data Science is Key to Achieving Energy Access in Africa | Madeleine Gleave | 26m |
Nithio builds an analytics + finance platform that uses CRM payments, IoT usage and high-resolution geospatial rasters to target off-grid solar customers in… | ||
| Scenario Analysis using Global Climate Models | Gopal Erinjippurath | 26m |
Quantifies annual, asset-level climate risk from 2020–2100 using ensembles from CMIP6 and scenario pathways SSP5-8.5 and SSP2-4.5, mapping wildfire, inland… | ||
| Exploring Value of Synthetic Data from Overhead Perspective | Jacob Shermeyer | 20m |
RarePlanes is an open dataset combining 14,700 hand-annotated aircraft from 253 Maxar WorldView-3 scenes with 50,000 synthetic images and ~630,000 aircraft… | ||
| Stop Explaining Black Box Models, Use Interpretable Models Instead | Dr. Rudin | 26m |
Argument: replace black boxes with interpretable models for high‑stakes decisions because explanations of black boxes are unreliable. Examples: COMPAS and… | ||
| ML for the Real World: Distributionally Robust Extrapolation | Anqi Angie Liu | 30m |
Introduces a distributionally robust minimax framework that gives formal extrapolation guarantees under covariate shift, constraining an adversary by source… | ||
| Learning with Limited Labels & Weak Supervision | Rishabh Mehrotra | 23m |
Combines thousands-scale weak labels (clicks, dwell time) with ~10k explicit judgments and shows concrete gains from submodular subset selection and… | ||
| Outwitting Deep Learning Models | Igor Markov | 30m |
Shows concrete deployments where data-driven regular expressions power real-time COVID-19 detection in 64 languages, run from SQL/Python/PHP, handle… | ||
| Deep Learning Battle: TensorFlow 2 vs. PyTorch | Jon Krohn | 32m |
Survey concludes that TensorFlow 2 prioritizes production: TensorFlow 2 integrates Keras, offers TensorFlow Serving, tfjs, and Lite for deployment, and… | ||
| Shparkley: Scaling Shapley values with Spark | Cristine Marsh, Isaac Joseph | 26m |
Implementation of Shapley IME in Spark scales exact Monte Carlo Shapley approximations to large training sets by broadcasting the model to executors… | ||
| Lessons Learned from Building a Real-World Knowledge Graph | John Maiden | 30m |
Built a temporal knowledge graph for commercial real estate to recover true property ownership and contact info, using public tax, transaction and permit feeds… | ||
| Predicting the Unpredictable Cruise’s Continuous Learning Machine | Tianshi Gao | 28m |
Cruise builds a Continuous Learning Machine that deploys a simple model, logs every 10 Hz prediction, finds surprising mistakes (U‑turns, lane cuts… | ||
| Adversarial Examples, Defense Methods for Online Fraud Detection | Nitin Sharma | 30m |
Presents an ensemble adversarial generator that combines gradient methods (fast gradient sign and iterative), iterative optimization (Carlini Wagner)… | ||
| Demand Sensing Decision Support System | Kriti Kohli | 23m |
Claims a deployed demand‑sensing system on IBM Cloud that combines a curated data lake, NLP on social and news text, market and public‑health feeds, and client… | ||
| DevOps for Data Science With Kubernetes | Sophie Watson | 28m |
Demonstrates a pipeline that takes Git-hosted Jupyter notebooks, uses a source-to-image builder to extract dependencies, run training and produce a… | ||
| Efficient BERT: Optimal Multimetric Bayesian Optimization | Meghana Ravikumar | 24m |
Distillation of BERT pairs with SigOpt multimetric Bayesian optimization and Ray on 20 EC2 instances to search architectures and hyperparameters jointly… | ||
| Startup Showcase | John Whaley, Mohan Reddy | 39m |
Four startups show concrete wins not vague promises: UnifyID claims passive, on‑device authentication deployed on 34 million devices with ~1 in 50,000… | ||
| Unique Challenges of AI in Materials Design | Alexandra Johnson | 22m |
Materials design faces three concrete obstacles and specific fixes: messy handwritten lab data is tamed by a formulation-aware data platform that preserves… | ||
| Optimized Routing with Satellite Imagery and Computer Vision | Adam Van Etten | 16m |
Claims an end-to-end system called CRESI baseline that extracts road graphs plus per-segment speeds from satellite imagery and uses them for true time-optimal… | ||
| Developing and Delivering Personalized Polygenic Scores at Scale | James Ashenhurst, Shannon Hamilton | 29m |
An end-to-end pipeline turns GWAS summary statistics from ~1.8M training samples into candidate polygenic scores evaluated on ~200k validation and similar test… | ||
| Adversarial Fairness in the Labor Market | Patrick Hagerty | 19m |
Claims a production method called Adversarial Fairness that trains a KPI predictor against an adversary that tries to infer EEOC protected classes, optimizing… | ||
| Still using one hot encoding? Here’s some ‘HELP’! | Vinay Prabhu | 28m |
Proposal replaces one-hot labels with dense, lower-dimensional prototypes computed by equiangular Grassmannian line packing, trains networks as regression… | ||
| Shopping Recommendations | Somnath Banerjee | 21m |
Pinterest turns a query product into a few hundred candidates from a 100M+ catalog by computing product embeddings with a graph convolutional network over the… | ||
| nothing matches | ||