The power of comparative reasoning

By Jay Yagnik, Dennis W. Strelow, David A. Ross, Ruei-Sung Lin. 2011

141 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Seeing 3D Chairs: Exemplar Part-Based 2D-3D Alignment Using a Large Dataset of CAD Models (2014, methodology): “For instance, the recent part-based hashing technique of [10] could be directly applied to our algorithm by applying winner-take-all hashing [40] on the discriminative visual elements.”
Automatically Discovering and Learning New Visual Categories with Ranking Statistics (2020, background): “Finally, other works (Dean et al., 2013; Yagnik et al., 2011) discuss the application of rank statistics to measuring the similarity of vectors; however, to the best of our knowledge, we are the ﬁrst to apply rank statistics to the task of novel category discovery using deep neural networks.”
Speeding Up Neural Networks for Large Scale Classification using WTA Hashing (2015, background): “In the resulting space Hamming distance closely correlates with rank similarity measure [11].”
Novel Visual Category Discovery with Dual Ranking Statistics and Mutual Knowledge Distillation (2021, background): “It is proven that ranking statistics is robust to noise, especially in high-dimensional space [51].”
Optimal Data-Dependent Hashing for Approximate Near Neighbors (2015, background): “Many or most of the practical applications of LSH involve designing data-aware hash functions [27, 19, 29, 31, 26, 32] (see also a recent survey [30]).”
Discriminative Orderlet Mining for Real-Time Recognition of Human-Object Interaction (2014, methodology): “Second, the order features are usually randomly generated in the previous work [3] while we use a data mining process to select the discriminative orderlets.”
Fast, Accurate Detection of 100,000 Object Classes on a Single Machine (2013, methodology): “We allow filters to operate on diverse features and achieve additional benefits by embedding patches of the resulting feature map in an ordinal space using a highdimensional sparse feature descriptor [21].”
Deep Networks With Large Output Spaces (2014, methodology): “In this work, we employ the subfamily of hash functions, winner-take-all (WTA) hashing introduced in (Yagnik et al., 2011), since it has been successfully applied for the similar task of scaling up ﬁlter-based object detection in (Dean et al., 2013).”
Kinome‐Wide Profiling Prediction of Small Molecules (2018, methodology): “The length of the target descriptor was adjusted to 2048 using a WTA-hash function to match the length of the compound fingerprint.([38]) The length of the kinase ID fingerprint was adjusted to that of the chemical fingerprint by simple repetition.”
A Training-free Classification Framework for Textures, Writers, and Materials (2012, methodology): “The power of comparative reasoning (WTA hash [38]) enhances and robustifies the representations by adding resilience to perturbations in numeric values.”
Hashing for Similarity Search: A Survey (2014, methodology): “Winner Take All (WTA) hash [140] is a sparse embedding method that transforms the input feature space into binary codes such that the Hamming distance in the resulting space closely correlates with rank similarity measure.”
Joint Representation Learning and Novel Category Discovery on Single- and Multi-modal Data (2021, methodology): “As discussed in [48], the precise values in the highdimensional embedding is often not important, and the relative magnitude matters more.”