MLconf 2022 San Francisco
who it's for
ml engineers, ml platform engineers, data scientists, data engineers, engineering managers, research scientists, product managers
topics
invoice image character-level extraction; machine-readable AI factsheets governance; AutoML pipeline generation for tabular; transformer multitask taxonomy linking; LiDAR camera radar autonomous stacks; metadata-backed traceable ML platforms; feature stores and model registries; federated learning attacks and defenses; few-shot learning with large LMs; causal ML and contextual bandits
| title | author | length |
|---|---|---|
| Using Deep Learning to Understand Documents | Eitan Anzenberg | 42m |
An end-to-end computer-vision model extracts key fields directly from invoice images, trained on millions of images and using a convolutional encoder + GRU… | ||
| AI Factsheets as Industry Standard | Armand Ruiz | 23m |
IBM’s AI Fact Sheets system creates machine-readable reports that capture metadata automatically by a one-line library call and populate industry and… | ||
| A Generative AutoML for Tabular Data | Mehdi Bahrami | 35m |
A Fujitsu AutoML system generates executable machine‑learning pipelines for tabular data, outputs Python and Jupyter source code, and attaches line‑by‑line… | ||
| Industrial Applications of Machine Learning in Search | Jay Wang | 35m |
Microsoft Newsfeed content understanding builds hierarchical content signals from title, snippet and multimedia using a pre-trained Transformer multitask model… | ||
| Driving Autonomous Vehicles Forward w/Real World Applications | Vinutha Kallem | 27m |
A deployed urban AV stack that combines LiDAR, cameras and radar with a fifth‑generation "wayo driver" claims 20 million autonomous miles and 20 billion… | ||
| Empowering Traceable & Auditable ML in Production | Hendrix, Jonathan Jin | 27m |
Spotify runs a centralized ML platform called Hendrix that is built for traceability and auditability and whose core claim is a unifying Hendrix Ontology plus… | ||
| Essential Ingredients in Scaling Organizations for ML | Ali Arsanjani | 30m |
Argues that scaling ML requires operational components not bigger models alone: cites PaLM 540B trained on two Cloud TPU v4 pods (~6,000 chips) and Minerva… | ||
| Humans helping machines help humans run machines | Leah Mcguire | 25m |
Benchling builds an automated pipeline that produces usable models on very small, custom tabular growth datasets (far below the ~3,000 rows where deep learning… | ||
| Balancing Thin Line Between Data Intelligence & Privacy | Dr. Sherin Mathews | 29m |
Systematic overview claims Federated learning lets models train across isolated data while keeping raw data local and lists concrete attacks: inference… | ||
| Few-Shot Learning in Conversational AI | Mahnoosh Mehrabani | 26m |
Shows that humans hit >60% accuracy with just five examples on 3x3 rule tasks while standard ML (decision tree, logistic, MLP) lags and sometimes never learns… | ||
| Building Multi-Tenant Compute Systems in the Enterprise | Zachary Hanif | 44m |
Capital One moves dozens of siloed teams toward a single multi-tenant ML compute stack, standardizing on Kubeflow for workflows, using Rubicon for experiment… | ||
| Scaling up Lyft’s Growth via Causal ML and Bandits | Lei Tang | 31m |
Lyft builds a growth stack that predicts and optimizes incremental impact: HTE models estimate per-user lift from coupons, a linear program allocates a budget… | ||
| Semantic Label Representation and Multimodal Categorization | Binwei Yang | 32m |
Product categorization uses CLIP embeddings to get category name vectors, computes cosine similarities, and truncates to top-20 nearest neighbors to build soft… | ||
| nothing matches | ||