Performance Comparison of Machine Learning Platforms

Roy, Asim; Qureshi, Shiban; Pande, Kartikeya; Nair, Divitha; Gairola, Kartik; Jain, Pooja; Singh, S.R.K.; Sharma, Kirti; Jagadale, Akshay; Lin, Yi; Sharma, Shashank; Gotety, Ramya; Zhang, Yuexin; Ji, Tang; Mehta, Tejas; Sindhanuru, Hemanth; Okafor, Nonso; Das, S.; Gopal, Chidambara N.; Rudraraju, Srinivasa B.; Kakarlapudi, Avinash V.

doi:10.1287/ijoc.2018.0825

Cited by 15 publications

(10 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our work extends previous efforts to obtain actionable knowledge for researchers and ML through benchmark studies of ML algorithms (Baumann et al, 2019;Fitzpatrick and Mues, 2016;Gartner et al, 2015) and ML platforms (Roy et al, 2019). Thus, we seek to address the lack of an overview to effective FSMs in particular problem classes.…”

Section: Objectives Of This Workmentioning

confidence: 84%

Filter Methods for Feature Selection in Supervised Machine Learning Applications -- Review and Benchmark

Hopf¹,

Sascha²

2021

Preprint

View full text Add to dashboard Cite

The amount of data for machine learning (ML) applications is constantly growing. Not only the number of observations, especially the number of measured variables (features) increases with ongoing digitization. Selecting the most appropriate features for predictive modeling is an important lever for the success of ML applications in business and research. Feature selection methods (FSM) that are independent of a certain ML algorithm-so-called filter methods-have been numerously suggested, but little guidance for researchers and quantitative modelers exists to choose appropriate approaches for typical ML problems. This review synthesizes the substantial literature on feature selection benchmarking and evaluates the performance of 58 methods in the widely used R environment. For concrete guidance, we consider four typical dataset scenarios that are challenging for ML models (noisy, redundant, imbalanced data and cases with more features than observations). Drawing on the experience of earlier benchmarks, which have considered much fewer FSMs, we compare the performance of the methods according to four criteria (predictive performance, number of relevant features selected, stability of the feature sets and runtime). We found methods relying on the random forest approach, the double input symmetrical relevance filter (DISR) and the joint impurity filter (JIM) were well-performing candidate methods for the given dataset scenarios.

show abstract

Section: Objectives Of This Workmentioning

confidence: 84%

Filter Methods for Feature Selection in Supervised Machine Learning Applications -- Review and Benchmark

Hopf¹,

Sascha²

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…This means that while we consider startups where AI is an enabler or outcome of the business model, we do not consider digital startups as AI startups where AI technology merely acts as the context to improve its work processes [16]. Examples of AI startups include companies in categories such as machine learning [28][29][30], intelligent systems [31][32][33], natural language processing [34][35][36][37][38], and predictive analytics [39][40][41][42].…”

Section: Growth Of Ai Startupsmentioning

confidence: 99%

A Scaling Perspective on AI Startups

Schulte-Althoff¹,

Fürstenau²,

Lee³

2021

Proceedings of the Annual Hawaii International Conference on System Sciences

View full text Add to dashboard Cite

Digital startups' use of AI technologies has significantly increased in recent years, bringing to the fore specific barriers to deployment, use, and extraction of business value from AI. Utilizing a quantitative framework regarding the themes of startup growth and scaling, we examine the scaling behavior of AI, platform, and service startups. We find evidence of a sublinear scaling ratio of revenue to age-discounted employment count. The results suggest that revenueemployee growth pattern of AI startups is close to that of service startups, and less so to that of platform startups. Furthermore, we find a superlinear growth pattern of acquired funding in relation to the employment size that is largest for AI startups, possibly suggesting hype tendencies around AI startups. We discuss implications in the light of new economies of scale and scope of AI startups related to decisionmaking and prediction.

show abstract

“…This is done so that the method used is in accordance with the characteristics of the system because each system has different characteristics. Asim Roy et al [8] have compared several methods and evaluated machine learning algorithms based on performance measures (e.g., Accuracy, Area Under the Curve (AUC), and F-score). Such a method can be used to compare standard Machine Learning platforms such as SAS, IBM SPSS, and Microsoft Azure ML.…”

Section: Literature Reviewmentioning

confidence: 99%

Improved Performance of Trash Detection and Human Target Detection Systems using Robot Operating System (ROS)

Kisron

Dewantara

Oktavianto

2021

JRE

View full text Add to dashboard Cite

In a visual-based real detection system using computer vision, the most important thing that must be considered is the computation time. In general, a detection system has a heavy algorithm that puts a strain on the performance of a computer system, especially if the computer has to handle two or more different detection processes. This paper presents an effort to improve the performance of the trash detection system and the target partner detection system of a trash bin robot with social interaction capabilities. The trash detection system uses a combination of the Haar Cascade algorithm, Histogram of Oriented Gradient (HOG) and Gray-Level Coocurrence Matrix (GLCM). Meanwhile, the target partner detection system uses a combination of Depth and Histogram of Oriented Gradient (HOG) algorithms. Robotic Operating System (ROS) is used to make each system in separate modules which aim to utilize all available computer system resources while reducing computation time. As a result, the performance obtained by using the ROS platform is a trash detection system capable of running at a speed of 7.003 fps. Meanwhile, the human target detection system is capable of running at a speed of 8,515 fps. In line with the increase in fps, the accuracy also increases to 77%, precision increases to 87,80%, recall increases to 82,75%, and F1-score increases to 85,20% in trash detection, and the human target detection system has also improved accuracy to 81%, %, precision increases to 91,46%, recall increases to 86,20%, and F1-score increases to 88,42%.

show abstract

Performance Comparison of Machine Learning Platforms

Cited by 15 publications

References 48 publications

Filter Methods for Feature Selection in Supervised Machine Learning Applications -- Review and Benchmark

Filter Methods for Feature Selection in Supervised Machine Learning Applications -- Review and Benchmark

A Scaling Perspective on AI Startups

Improved Performance of Trash Detection and Human Target Detection Systems using Robot Operating System (ROS)

Contact Info

Product

Resources

About