RapidScorer

Ye, Ting; Zhou, Hucheng; Zou, Will Y.; Gao, Bin; Zhang, Ruofei

doi:10.1145/3219819.3219857

Cited by 16 publications

(5 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Modern boosting systems such as XGBoost and LightGBM (Ke et al, 2017) all have performant GP-GPU implementations (Mitchell and Frank, 2017;Zhang et al, 2017). Not only learning, but also storage and use of decision trees is optimized, for instance using bit-level data structures, to allow deployment on edge devices with limited resources (Lucchese et al, 2017;Ye et al, 2018;Koschel et al, 2023).…”

Section: Parallelization and Distributionmentioning

confidence: 99%

Decision trees: from efficient prediction to responsible AI

Blockeel¹,

Devos²,

Frénay‬³

et al. 2023

Front. Artif. Intell.

View full text Add to dashboard Cite

This article provides a birds-eye view on the role of decision trees in machine learning and data science over roughly four decades. It sketches the evolution of decision tree research over the years, describes the broader context in which the research is situated, and summarizes strengths and weaknesses of decision trees in this context. The main goal of the article is to clarify the broad relevance to machine learning and artificial intelligence, both practical and theoretical, that decision trees still have today.

show abstract

Section: Parallelization and Distributionmentioning

confidence: 99%

Decision trees: from efficient prediction to responsible AI

Blockeel¹,

Devos²,

Frénay‬³

et al. 2023

Front. Artif. Intell.

View full text Add to dashboard Cite

show abstract

“…The concept of vectoring the tree structures is also applied to the context of ranking models in [9], which enhances the QuickScorer algorithm for gradient boosted trees [5,8]. Ye et al in [13] further improve the scalability of such vectorization methods by encoding the node representation to compact the memory footprint. These techniques decompose the tree-ensembles into different data structures based on the feature values, which is especially effective for large ensembles of smaller trees.…”

Section: Related Workmentioning

confidence: 99%

Immediate Split Trees: Immediate Encoding of Floating Point Split Values in Random Forests

Hakert¹,

Chen²,

Chen³

2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Random forests and decision trees are increasingly interesting candidates for resource-constrained machine learning models. In order to make the execution of these models efficient under resource limitations, various optimized implementations have been proposed in the literature, usually implementing either native trees or if-else trees. While a certain motivation for the optimization of if-else trees is to benefit the behavior of dedicated instruction caches, in this work we highlight that if-else trees might also strongly depend on data caches. We identify one crucial issue of if-else tree implementations and propose an optimized implementation, which keeps the logic tree structure untouched and thus does not influence the accuracy, but eliminates the need to load comparison values from the data caches. Experimental evaluation of this implementation shows that we can greatly reduce the amount of data cache misses by up to ≈ 99%, while not increasing the amount of instruction cache misses in comparison to the state-of-the-art. We additionally highlight various scenarios, where the reduction of data cache misses draws important benefit on the allover execution time.

show abstract

“…Rather than creating balanced trees that minimize the depth of every path, it may make more sense to minimize the depth of the most likely paths. Additionally, as with (Lucchese et al, 2015b;Ye et al, 2018), inference cost may be a non-trivial function of the tree topology, where the tree depth may not be neatly correlated with efficiency.…”

Section: Mixed Optimization Strategies Of Inference Efficiencymentioning

confidence: 99%

“…More recently, Gil-Costa et al (2022) and Molina et al (2021) propose a novel design of of the QuickScorer algorithm and the application of binning or quantization techniques to tree ensembles to fully leverage novel, energy-efficient field-programmable gate arrays (FPGAs). Ye et al (2018) take the data structure in QuickScorer and make it more compact in their algorithm, RapidScorer. The first observation was that nodemasks are two sequences of 1's separated by a sequence of 0's (i.e., 1 a 0 b 1 c for some a, b, c ≥ 0), and that only the sequence 0 b is relevant for the logical AND operations.…”

Section: Feature-major Traversalmentioning

confidence: 99%

“…Following these observations, Ye et al (2018) propose a more compact representation of nodemasks; a merging mechanism to store and execute repeated node tests only once; and a suitable data structure to allow efficient use of SIMD instructions that operate on 256 (or larger) bit-wide registers. Together this added compactness and parallelism reduces the algorithm's memory footprint and number of operations.…”

Section: Feature-major Traversalmentioning

confidence: 99%

See 1 more Smart Citation

ReNeuIR: Reaching Efficiency in Neural Information Retrieval

Bruch¹,

Lucchese

Nardini

2022

Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval

View full text Add to dashboard Cite

As information retrieval researchers, we not only develop algorithmic solutions to hard problems, but we also insist on a proper, multifaceted evaluation of ideas. The literature on the fundamental topic of retrieval and ranking, for instance, has a rich history of studying the effectiveness of indexes, retrieval algorithms, and complex machine learning rankers, while at the same time quantifying their computational costs, from creation and training to application and inference. This is evidenced, for example, by more than a decade of research on efficient training and inference of large decision forest models in Learning to Rank (LtR). As we move towards even more complex, deep learning models in a wide range of applications, questions on efficiency have once again resurfaced with renewed urgency. Indeed, efficiency is no longer limited to time and space; instead it has found new, challenging dimensions that stretch to resource-, sampleand energy-efficiency with ramifications for researchers, users, and the environment.This monograph takes a step towards promoting the study of efficiency in the era of neural information retrieval by offering a comprehensive survey of the literature on efficiency and effectiveness in ranking, and to a limited extent, retrieval. This monograph was inspired by the parallels that exist between the challenges in neural network-based ranking solutions and their predecessors, decision forest-based LtR models, as well as the connections between the solutions the literature to date has to offer. We believe that by understanding the fundamentals underpinning these algorithmic and data structure solutions for containing the contentious relationship between efficiency and effectiveness, one can better identify future directions and more efficiently determine the merits of ideas. We also present what we believe to be important research directions in the forefront of efficiency and effectiveness in retrieval and ranking.

show abstract

RapidScorer

Cited by 16 publications

References 23 publications

Decision trees: from efficient prediction to responsible AI

Decision trees: from efficient prediction to responsible AI

Immediate Split Trees: Immediate Encoding of Floating Point Split Values in Random Forests

ReNeuIR: Reaching Efficiency in Neural Information Retrieval

Contact Info

Product

Resources

About