A
Ali Farhadi
Researcher at University of Washington
Publications - 247
Citations - 87076
Ali Farhadi is an academic researcher from University of Washington. The author has contributed to research in topics: Context (language use) & Question answering. The author has an hindex of 63, co-authored 234 publications receiving 57227 citations. Previous affiliations of Ali Farhadi include University of Illinois at Urbana–Champaign & Lorestan University of Medical Sciences.
Papers
More filters
Posted Content
ELASTIC: Improving CNNs with Instance Specific Scaling Policies.
TL;DR: This paper introduces ELASTIC, a simple, efficient and yet very effective approach to learn instance-specific scale policy from data, and forms the scaling policy as a non-linear function inside the network’s structure that is learned from data.
Proceedings ArticleDOI
A Task-Oriented Approach for Cost-Sensitive Recognition
TL;DR: This paper proposes a novel cost-sensitive task-oriented recognition method that is able to generalize to unseen tasks for which there is no training data and outperforms state-of-the-art cost-based recognition baselines on the authors' new task-based dataset.
Posted Content
NCAM: Near-Data Processing for Nearest Neighbor Search
TL;DR: To enable large-scale computer vision, a new class of associative memories called NCAMs are proposed which encapsulate logic with memory to accelerate k-nearest neighbors and can improve the performance of kNN by orders of magnitude.
Proceedings Article
Learning Neural Network Subspaces
TL;DR: In this paper, a single method and in a single training run is proposed to learn lines, curves, and simplexes of high-accuracy neural networks, which can be ensembled, approaching the ensemble performance of independently trained networks without the training cost.
Posted Content
Are We Overfitting to Experimental Setups in Recognition
Matthew Wallingford,Aditya Kusupati,Keivan Alizadeh-Vahid,Aaron Walsman,Aniruddha Kembhavi,Ali Farhadi +5 more
TL;DR: A new framework is constructed, FLUID, which removes certain assumptions made by current experimental setups while integrating these sub-tasks via the following design choices -- consuming sequential data, allowing for flexible training phases, being compute aware, and working in an open-world setting.