Machine Learning Services Research Articles

Ease.ml in action

We demonstrate ease.ml, a multi-tenant machine learning service we host at ETH Zurich for various research groups. Unlike existing machine learning services, ease.ml presents a novel architecture that supports multi-tenant, cost-aware model selection that optimizes for minimizing total regrets of all users. Moreover, it provides a novel user interface that enables declarative machine learning at a higher level: Users only need to specify the input/output schemata of their learning tasks and ease.ml can handle the rest. In this demonstration, we present the design principles of ease.ml, highlight the implementation of its key components, and showcase how ease.ml can help ease machine learning tasks that often perplex even experienced users.

MLbench

Modern machine learning services and systems are complicated data systems --- the process of designing such systems is an art of compromising between functionality , performance , and quality . Providing different levels of system supports for different functionalities, such as automatic feature engineering, model selection and ensemble, and hyperparameter tuning, could improve the quality, but also introduce additional cost and system complexity. In this paper, we try to facilitate the process of asking the following type of questions: How much will the users lose if we remove the support of functionality x from a machine learning service? Answering this type of questions using existing datasets, such as the UCI datasets, is challenging. The main contribution of this work is a novel dataset, MLBench, harvested from Kaggle competitions. Unlike existing datasets, MLBench contains not only the raw features for a machine learning task, but also those used by the winning teams of Kaggle competitions. The winning features serve as a baseline of best human effort that enables multiple ways to measure the quality of machine learning services that cannot be supported by existing datasets, such as relative ranking on Kaggle and relative accuracy compared with best-effort systems. We then conduct an empirical study using MLBench to understand example machine learning services from Amazon and Microsoft Azure, and showcase how MLBench enables a comparative study revealing the strength and weakness of these existing machine learning services quantitatively and systematically. The full version of this paper can be found at arxiv.org/abs/1707.09562

Ease.ml

We present ease.ml, a declarative machine learning service platform. With ease.ml, a user defines the high-level schema of an ML application and submits the task via a Web interface. The system then deals with the rest, such as model selection and data movement. The ultimate question we hope to understand is that, as a "service provider" that manages a shared cluster of machines running machine learning workloads, what is the resource sharing strategy that maximizes the global satisfaction of all our users?

RETRACTED ARTICLE: Classification of sentence level sentiment analysis using cloud machine learning techniques

Cloud machine learning (CML) techniques offer contemporary machine learning services, with pre-trained models and a service to generate own personalized models. This paper presents a completely unique emotional modeling methodology for incorporating human feeling into intelligent systems. The projected approach includes a technique to elicit emotion factors from users, a replacement illustration of emotions and a framework for predicting and pursuit user’s emotional mechanical phenomenon over time. The neural network based CML service has better training concert and enlarged exactness compare to other large scale deep learning systems. Opinions are important to almost all human activities and cloud based sentiment analysis is concerned with the automatic extraction of sentiment related information from text. With the rising popularity and availability of opinion rich resources such as personal blogs and online appraisal sites, new opportunities and issues arise as people now, actively use information technologies to explore and capture others opinions. In the existing system, a segmentation ranking model is designed to score the usefulness of a segmentation candidate for sentiment classification. A classification model is used for predicting the sentiment polarity of segmentation. The joint framework is trained directly using the sentences annotated with only sentiment polarity, without the use of any syntactic or sentiment annotations in segmentation level. However the existing system still has issue with classification accuracy results. To improve the classification performance, in the proposed system, cloud integrate the support vector machine, naive bayes and neural network algorithms along with joint segmentation approaches has been proposed to classify the very positive, positive, neutral, negative and very negative features more effectively using important feature selection. Also to handle the outliers we apply modified k-means clustering method on the given dataset. It is used to cloud cluster the outliers and hence the label as well as unlabeled features is handled efficiently. From the experimental result, we conclude that the proposed system yields better performance than the existing system.

Statistical and Hybrid Methods Implemented in a Web Application for Predicting Reservoir Inflows during Flood Events

Reservoir management is a critical component of flood management, and information on reservoir inflows is particularly essential for reservoir managers to make real-time decisions given that flood conditions change rapidly. This study's objective is to build real-time data-driven services that enable managers to rapidly estimate reservoir inflows from available data and models. We have tested the services using a case study of the Texas flooding events in the Lower Colorado River Basin in November 2014 and May 2015, which involved a sudden switch from drought to flooding. We have constructed two prediction models: a statistical model for flow prediction and a hybrid statistical and physics-based model that estimates errors in the flow predictions from a physics-based model. The study demonstrates that the statistical flow prediction model can be automated and provides acceptably accurate short-term forecasts. However, for longer term prediction (2 h or more), the hybrid model fits the observations more closely than the purely statistical or physics-based prediction models alone. Both the flow and hybrid prediction models have been published as Web services through Microsoft's Azure Machine Learning (AzureML) service and are accessible through a browser-based Web application, enabling ease of use by both technical and nontechnical personnel.

Estimage: A mobile application in agriculture for image to object count estimation

A major problem in the agricultural field is accurately estimating the yield of produce. Typically, farmers must wait to measure their crops after harvest using manual mechanical equipment. There is value in having better methods of yield estimation based on data that can be captured with inexpensive technology in the field, such as a smartphone. We develop a smartphone application for the Android platform with access to a cloud-based machine learning (ML) service that can estimate the amount of crop on a bush or tree from an image. The development of an image-to-object-count estimation system called Estimage is presented. The Estimage system consists of an Android client application for user interaction, a PHP server application for request handling, an Octave program for image normalization, and an open-source ML software package called ilastik that applies a predictive model to an image. The functionality of the mobile client application is tested and the system satisfies all of the functional requirements and most of the non-functional requirements. The system is tested on the images of coins on a table, logs stacked in a pile, and blueberries on the bush. We show that the system is capable of accurately estimating the count of objects in the images that have objects of the same (size, colour, and shape) as that used to train the model. The system is good at distinguishing object pixels from the image background pixels provided the background is consistent with that used to train the model.

‘Digital tech’ and the public sector: what new role after public funding?

ABSTRACTInnovation scholars have long recognized entrepreneurship is ‘imitative’, whereas the commercialization of novelty is ‘innovative’. Thus they are highly distinctive skill-sets. Entrepreneurship, first, involves optimizing market sentiment for pure profit sometimes to the point of catastrophe and even fraudulence in many markets. These include: payment protection insurance (PPI) to ‘flash crashes’, automotive emission ‘defeat devices’, corporate bribery settlements, social media ‘hacking’, ‘fake news’ and a litany of other infractions and catastrophes. Innovation, by contrast, is more explorative and team-reliant. Even if patenting betrays the hope for commercialization on markets, patented innovation frequently fails. Some academic innovators even profess a preference for prizes over profits. Second, this means that collective ‘bonding’ among entrepreneurs, in the form of claimed ‘entrepreneurial ecosystems’, is often based on a single customer platform or as a supplier of a highly specialist type of ‘imitative’ service from identikit pizza chains to ‘me-too’ smartphone apps. Through the latter, fused with artificial intelligence some interactive machine-learning services have long-existed as ‘postsocial’ algorithms serving customers of, for example, investment banks in stock and currency markets. Finally, entrepreneurship is fundamentally competitive, individualistic and non-solidaristic, whereas ‘open innovation’ was born from the practices of ‘open science’ and the collegiate tradition of research. Accordingly, ‘entrepreneurial ecosystems’ can display more closure than RIS set-ups. This special issue explores aspects of these ecosystem platforms and their implications for emergent forms of urban and regional evolution in the near and nearly present future.

A Machine Learning Approach for Visual Recognition of Complex Parts in Robotic Manipulation

This research presents a method for the visual recognition of parts using machine learning services to enable the manipulation of complex parts. Robotic manipulation of complex parts is a challenging application due to the uncertainty of the parts’ positioning as well as the gripper's grasping instability. This instability is caused by the non-symmetrical and complex geometries that may result in a slightly variable orientation of the part after being grasped, which is outside the handling/assembly process tolerance.To compensate for this, a visual recognition approach is implemented via classifiers. Finally, a case study focusing on the manipulation of consumer goods is demonstrated and evaluated.

Leveraging Cloud-Based Predictive Analytics to Strengthen Audience Engagement

To grow their business and increase their audience, content distributors must understand the viewing habits and interests of content consumers. This typically requires solving tough computational problems, such as rapidly processing vast amounts of raw data from web sites, social media, devices, catalogs, and back-channel sources. Fortunately, today's content distributors can take advantage of the scalability, cost effectiveness, and pay-as-you-go model of the cloud to address these challenges. In this paper, we show content distributors how to use cloud technologies to build predictive analytic solutions. We examine architectural patterns for optimizing media delivery, and we discuss how to assess the overall consumer experience based on representative data sources. Finally, we present concrete implementations of cloud-based machine learning services and show how to use the services to profile audience demand, to cue content recommendations, and to prioritize the delivery of related media.