Early exit optimizations for additive machine learned ranking systems

By B. B. Cambazoglu, Hugo Zaragoza, Olivier Chapelle, Jiang Chen, Ciya Liao, Zhaohui Zheng, Jon Degenhardt. 2010

161 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Networks and Communications, Management Science and Operations Research

How researchers see this paper

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

In Artificial Intelligence

In Computer Networks and Communications

In Management Science and Operations Research

Citations in the corpus

Multi-Stage Document Ranking with BERT (2019, background): “For certain (easy) queries, stages of the pipeline can be skipped entirely, known as “early exits” (Cambazoglu et al., 2010).”
DeeBERT: Dynamic Early Exiting for Accelerating BERT Inference (2020, positive): “Our work is inspired by Cambazoglu et al. (2010), Teerapittayanon et al. (2017), and Huang et al. (2018), but mainly differs from previous work in that we focus on improving model efﬁciency with minimal quality degradation.”
Effectiveness/efficiency tradeoffs for candidate generation in multi-stage retrieval architectures (2013, background): “By fixing the feature generation and reranking stages, we are able to isolate the end-to-end effectiveness and efficiency implications of different algorithms.”
Efficient Cost-Aware Cascade Ranking in Multi-Stage Retrieval (2017, background): “1 INTRODUCTION Learning-to-Rank (LTR) systems are now commonly deployed by major search engine companies and they have been repeatedly shown to be highly eective for a variety of search related problems [6, 15, 26, 30].”
Training Efficient Tree-Based Models for Document Ranking (2013, methodology): “Similarly, we also consider strategies such as early exits [21]”
The Greedy Miser: Learning under Test-time Budgets (2012, methodology): “Among all methods of early-exit the authors suggested, we plot the best performing one (Early Exit Using Proximity Threshold).”
Quality versus efficiency in document scoring with learning-to-rank models (2016, methodology): “…based on forests of regression trees may generate thousands of trees to be evaluated on hundreds of features modeling a single query-document pair, in order to predict scores used to effectively rank all the candidate documents for a given query ( Cambazoglu et al., 2010; Segalovich, 2010 ).”
Fine-Tuning LLaMA for Multi-Stage Text Retrieval (2023, background): “While multi-stage retrieval pipelines date back well over a decade (Matveeva et al., 2006; Cambazoglu et al., 2010; Wang et al., 2011), they have bene-”
A Dual Embedding Space Model for Document Ranking (2016, background): “However, this is not a significant deficiency with the DESM as telescoping is a common initial set in industrial IR pipelines [7].”
Confident Adaptive Language Modeling (2022, background): “This approach, often referred to as early-exiting [16; 25; 47; 74; 79; 87], is complementary to many of the solutions above and can potentially be combined with them.”
Anytime Ranking for Impact-Ordered Indexes (2015, background): “Along the same lines, Cam-bazoglu et al. [6] introduced early-exit optimizations for ensembles of machine-learned rankers.”
A cascade ranking model for efficient ranked retrieval (2011, background): “There are many ways to prune documents based on such global information: both rank-based [8] and score-based pruning methods [8, 1] have been proposed in the past.”
Query Driven Algorithm Selection in Early Stage Retrieval (2017, background): “Multi-stage retrieval has become the dominant model in modern web search systems [3, 4, 9, 36, 37, 40].”