Downloads & Free Reading Options - Results

The revolution of Federated Learning - And we're back with another episode of the podcast When Machine Learning meets Privacy! For the episode #8 we've invited Ramen Dutta, a member of our community and founder of TensoAI. // Abstract: In this episode,  Ramen explain us the concept behind Federated Learning, all the amazing benefits and it's applications in different industries, particularly in agriculture. It's all about not centralizing the data, sound awkward? Just listen to the episode. //Other links to check on Ramen: https://www.linkedin.com/in/tensoai/ https://www.tensoai.com/ https://twitter.com/tensoAI //Final thoughts Feel free to drop some questions into our slack channel (https://go.mlops.community/slack)  Watch some of the other podcast episodes and old meetups on the channel: https://www.youtube.com/channel/UCG6qpjVnBTTT8wLGBygANOQ ----------- Connect With Us ✌️------------- Join our Slack community:   https://go.mlops.community/slack Follow us on Twitter:   @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Connect with Fabiana on LinkedIn: https://www.linkedin.com/in/fabiana-clemente/ Connect with Ramen on LinkedIn: https://www.linkedin.com/in/tensoai/

**AI and ethical dilemmas** Artificial Intelligence is seen by many as a vehicle for great transformation, but for others, it still remains a mystery, and many questions remain unanswered: will AI systems rule us one day? Can we trust AI to rule our criminal systems? Maybe create political campaigns and dominate political advertisements? Or maybe something less harmful, do our laundry? Some of these questions may sound absurd, but they are for sure making people shift from thinking purely about functional AI capabilities but also to look further to the ethics behind creating such powerful solutions.   For this episode we count with Charles Radclyffe as a guest, the data philosopher, to cover some of these dilemmas. You can reach out to Charles through LinkedIn or at ethicsgrade.io   Useful links:   - MLOps.Community slack - TEDx talk - Surviving the Robot Revolution - Digital Ethics whitepaper

MLOps Community Meetup #93! Two weeks ago, we talked to Chad Sanderson, Trustworthy Data for Machine Learning. //Abstract The most common challenge for ML teams operating at scale is data quality. In this talk, Chad discusses how Convoy invested in a large-scale data quality effort to treat data as an API and provide a data change management surface to enable trustworthy machine learning. // Bio Chad Sanderson is the Product Lead for Convoy's Data Platform team, which includes the data warehouse, streaming, BI & visualization, experimentation, machine learning, and data discovery. Chad has built everything from feature stores, experimentation platforms, metrics layers, streaming platforms, analytics tools, data discovery systems, and workflow development platforms. He's implemented open source, SaaS products (early and late-stage) and has built cutting-edge technology from the ground up. Chad loves the data space, and if you're interested in chatting about it with him, don't hesitate to reach out. // Related links    ----------- ✌️Connect With Us ✌️-------------    Join our Slack community:   https://go.mlops.community/slack Follow us on Twitter:   @mlopscommunity Sign up for the next meetup:   https://go.mlops.community/register Catch all episodes, Feature Store, Machine Learning Monitoring and Blogs: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Chad on LinkedIn: https://www.linkedin.com/in/chad-sanderson/

MLOps community meetup #64! Last Wednesday we talked to Christopher  Bergh, CEO, DataKitchen. //Abstract Working on a shared technically difficult problem there will be some things that are important no matter what industry you are in. Whether it's building cars in a factory, using agile or scrum methodology, or productionizing ML models you need a few basics. Chris gives us some of his best practices in the conversation. //Bio Chris Bergh is the CEO and Head Chef at DataKitchen. Chris has more than 25 years of research, software engineering, data analytics, and executive management experience. At various points in his career, he has been a COO, CTO, VP, and Director of Engineering. Chris is a recognized expert on DataOps. He is the co-author of the "DataOps Cookbook\" and the \"DataOps Manifesto,\" and a speaker on DataOps at many industry conferences. //Takeaways Your model is not an island. For success, Data science requires a high level of technical collaboration with other parts of the data organization. //Other Links On-Demand Webinar - Your Model is Not an Island:  Operationalizing Machine Learning at Scale with ModelOps   https://info.datakitchen.io/watch-on-demand-webinar-operationalize-machine-learning-at-scale-with-modelops ----------- Connect With Us ✌️-------------    Join our Slack community:  https://go.mlops.community/slack Follow us on Twitter:  @mlopscommunity Sign up for the next meetup:  https://go.mlops.community/register Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Chris on LinkedIn: https://www.linkedin.com/in/chrisbergh/ Timestamps: [00:00] Introduction to Christopher Bergh [02:57] MLOps community in partnership with MLOps World Conference [04:34] Chris' Background [07:59] "When we started with the company, I realized that the problem I have is generalizable to everyone. I'm getting enough there in years and I wanted to remove the amount of pain that other people have." [09:53] DataOps vs MLOps [10:15] "I don't really honestly care what Ops you use, right? Hahaha! Call it your favorite Ops 'cause first of all as an engineer, I want precise definitions. I look at it from a completely odd-ball way so you could call it whatever Ops term you want." [12:45] Best practices of companies [14:16] "When that code runs in production, monitor and check to see if it's right. Absorb it, monitor it because the model could go out of tune. The data going into it could be wrong. The data transformation could break. Shit happens and don't trust your data providers." [19:00] The whole is still greater than its part [20:26] "It is harder to focus on the results than just under a piece of the task. Don't spend too much time on doing the wrong thing." [23:50] DevOps Principles and Agile [27:17] DataOps Manifesto - DataOps is Data Management reborn [27:45] "The 'Ops' term is ending up encompassing the work that you do in addition to the system you build to do the work." [30:45] Standardization   [32:22] "I think that there's a lack of perception of the need to spend time on doing the operations part of the equation." [34:15] Tools as lego blocks [34:49] "Good interphases make good neighbors." [36:23] "Standards can help but they're not the panacea." [36:30] Cultural side - You build it, you own it, you ship it [39:28] Value chain [44:19] Ripple effect of testing [48:03] Google on "One tool to rule them all" [49:50] "Legacy happens if you're gonna live in the real world and not start greenfield projects." [53:47] Starting MLOps in the legacy system

Learn more about the technical aspects of artificial intelligence and machine learning by starting an ICFL Columbia University machine learning club. ICF Leaders empowers high school students to share their interests with other like-minded students around the globe.

We recently met up with Cormac Brick (Intel) and Mike Del Balso (Uber) at O'Reilly AI in SF. As the director of machine intelligence in Intel's Movidius group, Cormac is an expert in porting deep learning models to all sorts of embedded devices (cameras, robots, drones, etc.). He helped us understand some of the techniques for developing portable networks to maximize performance on different compute architectures. In our discussion with Mike, we talked about the ins and outs of Michelangelo, Uber's machine learning platform, which he manages. He also described why it was necessary for Uber to build out a machine learning platform and some of the new features they are exploring.

This is my second episode with David O'Hara, and as with the first episode I recorded with him, conversing with David is always a delight and a great pleasure. In this episode we discuss his trip to Central America, and we also talk about the recent archaeological discovery of a vast Mayan metropolis that "at its peak some 1,500 years ago, covered an area about twice the size of medieval England, with an estimated population of around five million." David goes over the cutting edge technology that is now being used to discover these, until very recently, hidden ruins of an ancient Mayan civilization, and what we can learn from these discoveries in regards to our own civilization. We also get into the ethics of artificial intelligence and the corporate control of the development of computer technology, and the implications this has for how information is disseminated through our society. David discusses some of the underlying issues on relying on algorithms and computer learning making big decisions for us, and how the kind of thinking leads to unintended outcomes. David O'Hara is a professor of Philosophy and Classics at Augustana University, and the author of the book "Downstream: Reflections on Brook Trout, Fly Fishing, and the Waters of Appalachia." David teaches a variety of courses on philosophy, classics, religion, and environmental ethics, and not long before the recording of this episode, had just recently got back from a trip to Central America where he teaches an in-depth course on reef ecology. Episode Notes:- Learn more about David's book "Downstream: Reflections on Brook Trout, Fly Fishing, and the Waters of Appalachia" here: https://wipfandstock.com/downstream.html- Read David's writings and get to understand him more at his blog: http://slowperc.blogspot.com/- Follow David on Twitter: https://twitter.com/Davoh- Read the article cited in this episode on the "Sprawling Maya network discovered under Guatemala jungle" here: http://www.bbc.com/news/world-latin-america-42916261- The song featured in this episode is "Sewee Sewee" by Mountain Man from the album Made The Harbor.- Podcast website: https://www.lastborninthewilderness.com- Support the podcast:PATREON: www.patreon.com/lastborninthewildernessONE-TIME DONATION: www.ko-fi.com/lastborninthewilderness- Follow and listen:SOUNDCLOUD: www.soundcloud.com/lastborninthewildernessITUNES: www.goo.gl/Fvy4caGOOGLE PLAY: https://goo.gl/wYgMQcSTITCHER: https://goo.gl/eeUBfS- Social Media:FACEBOOK: www.facebook.com/lastborninthewildernesspodcastTWITTER: www.twitter.com/lastbornpodcastINSTAGRAM: www.instagram.com/patterns.of.behavior

This week Elixir creator Jose Valim joins Jerod and Practical AI's Daniel Whitenack to discuss Numerical Elixir, his new project that's bringing Elixir into the world of machine learning. We discuss why Jose chose this as his next direction, the team's layered approach, influences and collaborators on this effort, and their awesome collaborative notebook project that's built on Phoenix LiveView.

Venture Capital in Machine Learning Startups With John Spindler CEO of Capital Enterprise.  John Spindler CEO of Capital Enterprise. We talked about what trends he has been seeing within MLOps, ML companies and also how he evaluates a deal.  John Spindler has over 15 years experience as an entrepreneur and business advisor/consultant and as well as being responsible for the day to day management of Capital Enterprise he is also a general partner at AI Seed, an early-stage fund that invests in highly talented AI-first companies.  John is on a mission to make it possible for someone moderately intelligent, with a good idea, ambition and passion to make it as an entrepreneur.  Join our slack community: https://join.slack.com/t/mlops-community/shared_invite/zt-391hcpnl-aSwNf_X5RyYSh40MiRe9Lw Follow us on twitter:@mlopscommunity Sign up for the next meetup: https://zoom.us/webinar/register/WN_a_nuYR1xT86TGIB2wp9B1g    Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with John on LinkedIn:  https://www.linkedin.com/in/johnbspindler/

Venture Capital in Machine Learning Startups With John Spindler CEO of Capital Enterprise.  John Spindler CEO of Capital Enterprise. We talked about what trends he has been seeing within MLOps, ML companies and also how he evaluates a deal.  John Spindler has over 15 years experience as an entrepreneur and business advisor/consultant and as well as being responsible for the day to day management of Capital Enterprise he is also a general partner at AI Seed, an early-stage fund that invests in highly talented AI-first companies.  John is on a mission to make it possible for someone moderately intelligent, with a good idea, ambition and passion to make it as an entrepreneur.  Join our slack community: https://join.slack.com/t/mlops-community/shared_invite/zt-391hcpnl-aSwNf_X5RyYSh40MiRe9Lw Follow us on twitter:@mlopscommunity Sign up for the next meetup: https://zoom.us/webinar/register/WN_a_nuYR1xT86TGIB2wp9B1g    Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with John on LinkedIn:  https://www.linkedin.com/in/johnbspindler/

Venture Capital in Machine Learning Startups With John Spindler CEO of Capital Enterprise.  John Spindler CEO of Capital Enterprise. We talked about what trends he has been seeing within MLOps, ML companies and also how he evaluates a deal.  John Spindler has over 15 years experience as an entrepreneur and business advisor/consultant and as well as being responsible for the day to day management of Capital Enterprise he is also a general partner at AI Seed, an early-stage fund that invests in highly talented AI-first companies.  John is on a mission to make it possible for someone moderately intelligent, with a good idea, ambition and passion to make it as an entrepreneur.  Join our slack community: https://join.slack.com/t/mlops-community/shared_invite/zt-391hcpnl-aSwNf_X5RyYSh40MiRe9Lw Follow us on twitter:@mlopscommunity Sign up for the next meetup: https://zoom.us/webinar/register/WN_a_nuYR1xT86TGIB2wp9B1g    Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with John on LinkedIn:  https://www.linkedin.com/in/johnbspindler/

Practical Mlops Operationalizing Machine Learning Models 

The rise in neurological disorders within modern society emphasizes the critical need for accurate diagnosis and immediate treatment. Traditional diagnostic methods mostly depend on specialized Neurologists and primarily utilize MRI scans and related neuroimaging techniques to evaluate the neurological health of patients. With today’s growing need for diagnosing neural disorders, we need more prominent automated diagnostic tools to enhance medical practices. In this system, we propose a novel approach to diagnosing neurological disorders such as Alzheimer's Disease, Brain tumor and Brain stroke using Sequential Convolutional Neural Networks (CNNs) and machine learning algorithms. Our system utilizes the CNN at most of its capabilities in analyzing neuroimaging data and extracting key features that indicate neurological disorders. Through rigorous training and validation processes, the system achieves notable accuracy in the identification and classification of neurological disorders based on neuroscan findings. This enhances healthcare delivery by improving patient outcomes in the field of neurology.

Online payment fraud poses a significant threat to financial transactions, resulting in substantial economic losses. This project proposes a machine learning-based system to predict and detect fraudulent transactions using the Decision Tree Classifier algorithm. Online payment fraud involves unauthorized access and manipulation of financial transactions, including identity theft, phishing, card skimming, and transaction tampering. The Decision Tree Classifier algorithm trains on historical transaction data (fraudulent and non-fraudulent) to build a classifier model. This model predicts transactions as fraudulent or non-fraudulent based on feature extraction and splitting. The process begins with user registration, where customers provide their bank details in the Website, which are then securely stored in an SQL database. Next, transaction details are input into the system, allowing for real-time monitoring. A Decision Tree Classifier-based machine learning model is employed to predict potential fraud, analyzing the collected data to identify patterns and anomalies. The prediction results are then displayed on the website, alerting users to potential fraud or confirming legitimate transactions. To ensure timely notification, an Email API is integrated, sending alerts to users and administrators when suspicious activity is detected. This comprehensive system provides a robust defense against online payment fraud, safeguarding users' financial information and maintaining trust in e-commerce transactions.

The need for personalized health solutions is increasing, and the fitness recommendation systems are mandatory to optimizing workout plans. This paper discusses a system providing personalized fitness recommendations based on Reinforcement Learning (RL), Long Short-Term Memory (LSTM) networks, Genetic Algorithms (GA), and Artificial Neural Networks (ANN). Such AI techniques would ensure real-time workout adjustments, future performance predictions, and optimization of fitness routine. Reinforcement Learning provides an adaptive workout planner and optimizes the user recommendations based upon user feedback that improves long-term health and fitness outcomes. LSTM Networks inherit time-series data such as past performance in workouts and are able to predict trends in fitness for the future and therefore offer training modifications ahead of time. Genetic Algorithms operate in selecting and mutating workout parameters so that the plans can dynamically evolve with constant adaptation. ANN improves testing patterns and mapping cause-effect relationships that make workouts more efficient. The evaluation is carried out on the architecture against applicable baselines and therefore shows the ability of the developed prototype to create adaptive fitness recommendations. The work demonstrates how many AI techniques combat the limitations of traditional fitness systems, such as lack of real-time adaptability and long-term optimization. The system produces a more intelligent approach in workout personalization through data to enhance the efficiency and satisfaction derived from fitness programs regarding a specific and dynamic approach to fitness planning.

Automated verification of software systems is a challenging problem because of their large (and often infinite) state-space. In this talk, we explore techniques from computational learning theory for verification of such systems. We show that learning can be effectively used to verify safety properties as well as liveness properties with fairness constraints. We can analyze both linear time and branching time temporal logics (more precisely omega-regular properties and Computational Tree Logic). We represent the states of the system as strings over some alphabet and use Angluin's algorithm for learning regular sets. We show that the learning based verification procedure is sound and, more interestingly, also complete if the fixpoints needed for verification are in fact regular. Finally, we conclude with discussion about a tool called LEVER which implements these techniques and some examples that we have analyzed using the tool. ©2005 Microsoft Corporation. All rights reserved.

Cancer immunotherapy has revolutionized cancer treatment by leveraging the immune system to target malignant cells, yet resistance in many cancers highlights the need for novel therapeutic targets. Artificial intelligence (AI) and machine learning (ML) have emerged as transformative tools for identifying new immunotherapy targets by analyzing vast datasets from genomics, proteomics, and clinical studies. This review explores the role of AI and ML in advancing the discovery of cancer-specific immunotherapy targets, such as tumor antigens and immune pathways. Key advances include the integration of big data, neoantigen prediction, biomarker discovery, and single-cell analysis. Despite these advancements, challenges remain, including data quality and standardization, interpretability of AI models, computational costs, and the need for biological validation of AI-driven discoveries. As AI and ML technologies continue to evolve, they hold the potential to overcome these barriers, leading to personalized immunotherapy solutions. This review also discusses future directions for AI-driven immunotherapy, emphasizing the need for improved models, ethical considerations, and clinical integration. 

In the era of cloud computing and large-scale distributed systems, ensuring uninterrupted service and operational reliability is crucial. Conventional fault tolerance techniques usually take a reactive approach, addressing problems only after they arise. This can result in performance deterioration and downtime. With predictive machine learning models, this research offers a proactive approach to fault tolerance for distributed systems, preventing significant failures before they arise. Our research focuses on combining cutting-edge machine learning algorithms with real-time analysis of massive streams of operational data to predict abnormalities in the system and possible breakdowns. We employ supervised learning algorithms such as Random Forests and Gradient Boosting to predict faults with high accuracy. The predictive models are trained on historical data, capturing intricate patterns and correlations that precede system faults. Early defect detection made possible by this proactive approach enables preventative remedial measures to be taken, reducing downtime and preserving system integrity. To validate our approach, we designed and implemented a fault prediction framework within a simulated distributed system environment that mirrors contemporary cloud architectures. Our experiments demonstrate that the predictive models can successfully forecast a wide range of faults, from hardware failures to network disruptions, with significant lead time, providing a critical window for implementing preventive measures. Additionally, we assessed the impact of these pre-emptive actions on overall system performance, highlighting improved reliability and a reduction in mean time to recovery (MTTR). We also analyse the scalability and adaptability of our proposed solution within diverse and dynamic distributed environments. Through seamless integration with existing monitoring and management tools, our framework significantly enhances fault tolerance capabilities without requiring extensive restructuring of current systems. This work introduces a proactive approach to fault tolerance in distributed systems using predictive machine learning models. Unlike traditional reactive methods that respond to failures after they occur, this work focuses on anticipating faults before they happen.

In today’s e-commerce in Nigeria, customers access online stores to browse through and place orders for products or services via the internet on their devices while some are skeptical due to the experiences from what I ordered versus what I got syndrome. Though this method of business has flourished to an extent, it greatly faces a crucial challenge in unravelling consumer’s sentiments particularly in the realm of product reviews. This deficiency inhibits most e-commerce platforms in Nigeria from gaining effective sensitivity into users’ preferences, thus, limiting their ability to boost their product recommendations and, understand and improve customers’ experiences. This research aims to bridge this gap by developing a sentiment analyzer of product in the e-commerce domain using Natural language processing and machine learning approach. The model will analyze the customers’ reviews based on positive or negative. The experimental data was collected from kaggle.com. Stemming and lemmatization were approaches used for cleaning the collected data. Features were extracted and transformed using CountVectorizer. Gaussian Naïve Bayes classifier was used as the machine learning technique. The model’s performance was evaluated and it returned 90% of accuracy, hence, an efficient and reliable model for product review sentiment analysis is developed.

In today's world, where speech recognition has become an integral part of our daily lives, the need for systems equipped with this technology has increased dramatically in the past few years. This research aims to locate the two selected Persian words in any given audio file. For this purpose, two standard and native datasets were prepared for this model one for train and the other for the test. Both datasets were converted into images of audio waveforms. Using the object detection technique, the model could extract different bounding boxes for each test audio, and then each box image goes through a CNN classifier and returns a corresponding label. Finally, a threshold is set so that only boxes with high accuracy are displayed as output. The results showed 93% accuracy for the CNN classifier and 50% accuracy for testing the model with object detection.

Introduction.  All polymer materials and composites based on them are characterized by pronounced rheological properties, the prediction of which is one of the most critical tasks of polymer mechanics. Machine learning methods open up great opportunities in predicting the rheological parameters of polymers. Previously, studies were conducted on the construction of predictive models using artificial neural networks and the CatBoost algorithm. Along with these methods, due to the capability to process data with highly nonlinear dependences between features, machine learning methods such as the  k -nearest neighbor method, and the support vector machine (SVM) method, are widely used in related areas. However, these methods have not been applied to the problem discussed in this article before. The objective of the research was to develop a predictive model for evaluating the rheological parameters of polymers using artificial intelligence methods by the example of polyvinyl chloride. Materials and Methods.  This paper used  k -nearest neighbor method and the support vector machine to determine the rheological parameters of polymers based on stress relaxation curves. The models were trained on synthetic data generated from theoretical relaxation curves constructed using the nonlinear Maxwell-Gurevich equation. The input parameters of the models were the amount of deformation at which the experiment was performed, the initial stress, the stress at the end of the relaxation process, the relaxation time, and the conditional end time of the process. The output parameters included velocity modulus and initial relaxation viscosity coefficient. The models were developed in the Jupyter Notebook environment in Python. Results.  New predictive models were built to determine the rheological parameters of polymers based on artificial intelligence methods. The proposed models provided high quality prediction. The model quality metrics in the SVR algorithm were: MAE – 1.67 and 0.72; MSE – 5.75 and 1.21; RMSE – 1.67 and 1.1; MAPE – 8.92 and 7.3 for the parameters of the initial relaxation viscosity and velocity modulus, respectively, with the coefficient of determination  R 2  – 0.98. The developed models showed an average absolute percentage error in the range of 5.9 – 8.9%. In addition to synthetic data, the developed models were also tested on real experimental data for polyvinyl chloride in the temperature range from 20° to 60°C. Обсуждение и вывод.   Апробация разработанных моделей на реальных экспериментальных кривых показала высокое качество их аппроксимации, сравнимое с другими методами. Таким образом, алгоритм k -ближайшего соседа и SVM могут использоваться для прогнозирования реологических параметров полимеров в качестве альтернативы искусственным нейронным сетям и алгоритму CatBoost, требуя меньших усилий для предварительной настройки. При этом в данном исследовании метод SVM оказался наиболее предпочтительным методом машинного обучения, поскольку он более эффективен при обработке большого количества признаков.  

Efficient and accurate data collection is paramount in clinical trials, and the design of Electronic Case Report Forms eCRFs plays a pivotal role in streamlining this process. This paper explores the integration of machine learning techniques in the design and implementation of eCRFs to enhance data collection efficiency. We delve into the synergies between eCRF design principles and machine learning algorithms, aiming to optimize data quality, reduce errors, and expedite the overall data collection process. The application of machine learning in eCRF design brings forth innovative approaches to data validation, anomaly detection, and real time adaptability. This paper discusses the benefits, challenges, and future prospects of leveraging machine learning in eCRF design for streamlined and advanced data collection in clinical trials. Dhanalakshmi D | Vijaya Lakshmi Kannareddy "Streamlining Data Collection: eCRF Design and Machine Learning" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-8 | Issue-1 , February 2024, URL: https://www.ijtsrd.com/papers/ijtsrd63515.pdf Paper Url: https://www.ijtsrd.com/biological-science/biotechnology/63515/streamlining-data-collection-ecrf-design-and-machine-learning/dhanalakshmi-d

A well ordered traffic management system is required in all types of roads, such as off roads, highways, etc. There has been several laws and speed controlled measures are taken in all places with different perspectives. Also Speed limit may vary from road to road. So there are number of methods has been proposed using computer Vision and machine learning algorithms for object tracking. Here vehicles are recognized and detected from the videos that taken using surveillance camera. The aim is to identification of the vehicles and tracking using Haar Classifier, then determine the speed of the vehicle and Finally Detecting the License plate of the vehicle. Detecting the License plate and vehicle speed using machine learning is tough but beneficial task. For the past few years Convolution Neural Network CNN has been widely used in computer vision for vehicle detection and identification. Dlibs are used to track the multiple objects at the same time. P. Devi Mahalakshmi | Dr. M. Babu "Automated License Plate detection and Speed estimation of Vehicle Using Machine Learning - Haar Classifier Algorithm" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-4 | Issue-6 , October 2020, URL: https://www.ijtsrd.com/papers/ijtsrd33395.pdf Paper Url: https://www.ijtsrd.com/engineering/computer-engineering/33395/automated-license-plate-detection-and-speed-estimation-of-vehicle-using-machine-learning--haar-classifier-algorithm/p-devi-mahalakshmi

Efficient tax debt collection is a challenge for Moroccan local tax authorities. This article explores the potential of machine learning techniques and novel strategies to enhance efficiency in this process. We present a practical use case demonstrating the application of machine learning for taxpayer segmentation, improving accuracy in identifying high-risk debtors. Using a comprehensive dataset of tax payment behavior, we showcase the effectiveness of machine learning algorithms in segmenting taxpayers based on their likelihood of non compliance or debt accumulation. We also investigate innovative strategies that integrate behavioral economics principles to enable better targeted interventions. Real-world case studies in local tax debt collection highlight the impact of these strategies. The findings underscore the transformative potential of machine learning techniques and novel strategies in improving the efficiency of local tax debt collection. Accurate identification of high-risk debtors and tailored enforcement actions help maximize revenue while minimizing resource waste. This research contributes to the existing knowledge by providing insights into the implementation of machine learning techniques and novel strategies in tax debt collection. It emphasizes the importance of data-driven approaches and highlights how local tax authorities can drive efficiency and optimize revenue collection by embracing these advancements.

Prediction of radionuclide diffusion enabled by missing data imputation and ensemble machine learning 作者： Tian, Mr. Junlei Feng, Mr. Jiaxing Shen, Mr. Jiacong Yao, Mr. Lei Wang, Mr. Jingyan WU, Dr. Tao zhao, Dr. Yaolin 赵耀林 通讯作者： WU, Dr. Tao Email:[email protected] 提交时间： 2025-01-12 17:30:56 摘要: Missing values in radionuclide diffusion datasets can undermine the predictive accuracy and robustness of machine learning (ML) models. A regression-based missing data imputation method using light gradient boosting machine (LGBM) algorithm was employed to impute over 60% of the missing data, establishing a radionuclide diffusion dataset containing 16 input features and 813 instances. The effective diffusion coefficient (De) was predicted using ten ML models. The predictive accuracy of ensemble meta-models, namely LGBM-extreme gradient boosting (XGB) and LGBM-categorical boosting (CatB), surpassed the other ML models, with R2 values of 0.94. The models were applied in predicting the De values of EuEDTA- and HCrO4- in saturated compacted bentonites at compaction ranged from 1200 kg/m3 to 1800 kg/m3, which was measured using a through-diffusion method. The generalization ability of LGBM-XGB model surpassed that of LGB-CatB in predicting the De of HCrO4-. Shapley additive explanations identified the total porosity as the most significant influencing factor. In addition, the partial dependence plot analysis technique showed clearer results for univariate correlation analysis. This study provides a regression imputation technique to refine radionuclide diffusion datasets, offering a deeper insight into analyzing the diffusion mechanism of radionuclide and supporting the safety assessment of the geological disposal of high-level radioactive waste. machine learning radionuclide diffusion bentonite regression imputation missing data diffusion experiments 来自： WU, Dr. Tao 分类： 物理学 >> 核物理学 备注: 已向《Nuclear Science and Techniques》投稿 引用： ChinaXiv:202501.00141 (或此版本 ChinaXiv:202501.00141V1 ) DOI:10.12074/202501.00141 CSTR:32003.36.ChinaXiv.202501.00141 推荐引用方式： Tian, Mr. Junlei,Feng, Mr. Jiaxing,Shen, Mr. Jiacong,Yao, Mr. Lei,Wang, Mr. Jingyan,WU, Dr. Tao,zhao, Dr. Yaolin 赵耀林.Prediction of radionuclide diffusion enabled by missing data imputation and ensemble machine learning.中国科学院科技论文预发布平台.[DOI:10.12074/202501.00141] 版本历史 [V1] 2025-01-12 17:30:56 ChinaXiv:202501.00141V1 下载全文

Prediction of radionuclide diffusion enabled by missing data imputation and ensemble machine learning 作者： Tian, Mr. Junlei Feng, Mr. Jiaxing Shen, Mr. Jiacong Yao, Mr. Lei Wang, Mr. Jingyan WU, Dr. Tao zhao, Dr. Yaolin 赵耀林 通讯作者： WU, Dr. Tao Email:[email protected] 提交时间： 2025-01-12 17:30:56 摘要: Missing values in radionuclide diffusion datasets can undermine the predictive accuracy and robustness of machine learning (ML) models. A regression-based missing data imputation method using light gradient boosting machine (LGBM) algorithm was employed to impute over 60% of the missing data, establishing a radionuclide diffusion dataset containing 16 input features and 813 instances. The effective diffusion coefficient (De) was predicted using ten ML models. The predictive accuracy of ensemble meta-models, namely LGBM-extreme gradient boosting (XGB) and LGBM-categorical boosting (CatB), surpassed the other ML models, with R2 values of 0.94. The models were applied in predicting the De values of EuEDTA- and HCrO4- in saturated compacted bentonites at compaction ranged from 1200 kg/m3 to 1800 kg/m3, which was measured using a through-diffusion method. The generalization ability of LGBM-XGB model surpassed that of LGB-CatB in predicting the De of HCrO4-. Shapley additive explanations identified the total porosity as the most significant influencing factor. In addition, the partial dependence plot analysis technique showed clearer results for univariate correlation analysis. This study provides a regression imputation technique to refine radionuclide diffusion datasets, offering a deeper insight into analyzing the diffusion mechanism of radionuclide and supporting the safety assessment of the geological disposal of high-level radioactive waste. machine learning radionuclide diffusion bentonite regression imputation missing data diffusion experiments 来自： WU, Dr. Tao 分类： 物理学 >> 核物理学 备注: 已向《Nuclear Science and Techniques》投稿 引用： ChinaXiv:202501.00141 (或此版本 ChinaXiv:202501.00141V1 ) DOI:10.12074/202501.00141 CSTR:32003.36.ChinaXiv.202501.00141 推荐引用方式： Tian, Mr. Junlei,Feng, Mr. Jiaxing,Shen, Mr. Jiacong,Yao, Mr. Lei,Wang, Mr. Jingyan,WU, Dr. Tao,zhao, Dr. Yaolin 赵耀林.Prediction of radionuclide diffusion enabled by missing data imputation and ensemble machine learning.中国科学院科技论文预发布平台.[DOI:10.12074/202501.00141] 版本历史 [V1] 2025-01-12 17:30:56 ChinaXiv:202501.00141V1 下载全文

The stock market is considered the primary domain of importance in the financial sector where Artificial Intelligence combined with various algorithmic practices empowers investors with datadriven insights, enhancing decision-making, predicting trends, and optimizing risk management for more informed and strategic financial outcomes. This research paper delves into the real-world applications of machine learning and algorithmic trading, observing their historical evolution together and how both of these can go hand in hand to control risk and forecast the movement of a stock or an index and its future. The research is structured to provide comprehensive insights into two major subdomains in the application of AI in algorithmic trading: risk management in equity markets and predictive analysis of stock trends through the application of machine learning models and training the current existing data which is feasible and training them with respect to historical scenarios of various market trends along with various fundamental and technical analysis techniques with the help of various deep learning algorithms. For risk management of a portfolio in finance, various machine learning models can be employed, depending on the specific needs and goals of the portfolio manager or risk analyst and implementing various valueat-risk algorithms along with deep learning techniques in order to assess risk at particular trade position and to manage volatile trades at unprecedented situations. The significance of this research paper lies in its practical applicability, offering real-world solutions to enhance trading strategies and decision-making processes with a focus on mitigating risk and capitalizing on market opportunities and also giving clear insights with respect to the current practical limitations of application of the provided solution and future scope to overcome the same.

Today, one of the top concerns for governments is road safety. There are many safety features built into cars, yet traffic accidents still happen frequently and are unavoidable. To lessen the harm caused by traffic accidents, predicting their causes has become the primary goal. In this situation, it will be beneficial to examine the frequency of accidents so that we can use this information to further aid us in developing strategies to lessen them. From this, we can deduce the connections between traffic accidents, road conditions, and the impact of environmental factors on accident occurrence. In order to construct an accident prediction model, I used machine learning techniques, including the Decision Tree, Random Forest, and Logistic Regression. The development of safety measures and accident prediction will both benefit from these classification systems. Several elements, including weather, vehicle condition, road surface condition, and light condition, can be used to predict road accidents. Three dataset files—accidents, casualties, and vehicles are loaded into this dataset. This allows us to forecast the severity of accidents.

Banks and other financial institutions compete for customers by providing a wide range of services and products. Most banks, however, make the vast majority of their money from their credit portfolio. Loans accepted by borrowers might lead to interest charges. The loan portfolio, and customers' repayment habits in particular, can have a substantial impact on a bank's bottom line. The financial institution's Non-Performing Assets can be reduced if it can accurately predict which borrowers are likely to default on their loans. Therefore, there is substantial scholarly value in exploring the prediction of loan endorsement. In order to make accurate predictions, it is crucial to use Machine Learning methods. Based on a person's past loan qualification history, this research uses a machine learning methodology to predict the person's likelihood of consistently making loan repayments. The primary aim of this research is to foretell how likely it is that a given individual will be granted a loan. 

CCC 044 / Machine Learning in Science and Engineering Machine Learning in Science and Engineering A Brief Introduction into Machine Learning with a few Application Examples A broad overview about the current stage of research in Machine Learning starting with the general motivation and the setup of learning problems and discussion of state-of-the-art learning algorithms for novelty detection, classification and regression. Additionally, machine learning methods used for spam detection, intrusion detection, brain computer interace and biological sequence analysis are outlined. The talk is going to have three parts: (a) What is Machine Learning about? This includes a general motivation, the setup of learning problems (suppervised vs unsupervised; batch vs online). I'll mention typical examples (e.g. OCR, Text-classification, medical Diagnosis, biological sequence analysis, time series prediction) and use them as motivation. (b) What are state-of-the-art learning techniques? With a minimal amount of theory, I'll describe some methods including a currently very successful and easily applicable method called Support Vector Machines. I'll provide references to standard literature and implementations of these algorithms. (c) I'll discuss a few applications in greater detail, to show how Machine Learning can be successfully applied in practice. These include: 1. spam detection 2. face detection and reconstruction 3. intelligent hard disk spin (online learning) 4. biological sequence analysis & drugs discovery 5. network intrusion detection 6. brain computer interface 7. analysis of questionnaires (Fraud detection, fake interviewer identification) I try not present the material as self-contained as possible, but I will require some math knowledge on part (b). I mainly want to bring ideas across and will provide references to papers and web-resources for further reading about the details of the methods and applications. More: http://www.ccc.de/congress/2004/fahrplan/event/44.en.html http://www.ccc.de/congress/2004/fahrplan/files/212-machine-learning-slides.pdf http://www.ccc.de/congress/2004/fahrplan/files/105-machine-learning-paper.pdf

Levy processes are random measures that give independent mass to independent increments. I will show how they can be used to model various types of data such as binary vectors or vectors of counts, with applications to text and images. These techniques fall in the category of nonparametric Bayesian methods, and are related to the better known Dirichlet process. ©2008 Microsoft Corporation. All rights reserved.

Rest easy with Google Play Protect. See how #Android’s machine learning systems make your device safer. #io17 https://t.co/3MjTIHxG2k Source: https://twitter.com/Android/status/864932679472566276 Uploader: Android

Application Of Machine Learning Techniques In Rice Leaf Disease Detection

30 July 2017 - DEF CON 25 (2017) Dan 'AltF4' Petro & Ben Morris - Bishop Fox https://www.bishopfox.com Weaponizing Machine Learning: Humanity Was Overrated Anyway /redirect?event=video_description&v=wbRx18VZlYA&redir_token=KZeUFQPTBuoWz6sabYG8Uy-f4yd8MTUyNjc1MzM4OUAxNTI2NjY2OTg5&q=https%3A%2F%2Fwww.defcon.org%2Fhtml%2Fdefcon-25%2Fdc-25-speakers.html%23Petro At risk of appearing like mad scientists, reveling in our latest unholy creation, we proudly introduce you to DeepHack: the open-source hacking AI. This bot learns how to break into web applications using a neural network, trial-and-error, and a frightening disregard for humankind. DeepHack can ruin your day without any prior knowledge of apps, databases - or really anything else. Using just one algorithm, it learns how to exploit multiple kinds of vulnerabilities, opening the door for a host of hacking artificial intelligence systems in the future. This is only the beginning of the end, though. AI-based hacking tools are emerging as a class of technology that pentesters have yet to fully explore. We guarantee that you'll be either writing machine learning hacking tools next year, or desperately attempting to defend against them. No longer relegated just to the domain of evil geniuses, the inevitable AI dystopia is accessible to you today! So join us and we'll demonstrate how you too can help usher in the destruction of humanity by building weaponized machine learning systems of your own - unless time travelers from the future don't stop us first. Source: https://www.youtube.com/watch?v=wbRx18VZlYA Uploader: Bishop Fox

Werbematerial des Buches "Machine Learning – Medien, Infrastrukturen und Technologien der Künstlichen Intelligenz", herausgegeben von Christoph Engemann und Andreas Sudmann. Veröffentlicht wurden auf der Website des Transcript-Verlages das Cover des Buches, eine Leseprobe und ein MARC-Report, alles kostenlos zum Download.

Non-linear loads can cause transients in electronic switches. They also result in a fluctuating output when the device is switched ON or OFF. These transients can harm not only the switches but also the devices that they are connected to, by passing excess currents or voltages to the devices. By applying machine learning, we can improve the gate drive voltages of the switches and thereby reduce switch transients. A feedback system is built that measures the output transients and then feeds it to a neural network algorithm that then gives a proper gate drive to the device. This will reduce transients and also improve performances of switch based devices like inverters and converters.

Predicting structural performance is a critical aspect of civil engineering, ensuring the safety, efficiency, and durability of buildings and infrastructure. Traditional methods, such as finite element analysis and empirical modeling, often fall short in addressing the complexities of modern structural systems. The advent of machine learning (ML) has revolutionized this domain by offering data-driven approaches capable of handling non-linear relationships and large datasets, enhancing the accuracy and efficiency of structural performance predictions. This review paper examines the applications of ML techniques, including Artificial Neural Networks (ANN), Support Vector Regression (SVR), Random Forest (RF), Decision Tree Regression (DTR), and hybrid models, in predicting structural metrics such as load-bearing capacity, deflection, durability, and seismic performance. The paper synthesizes findings from recent studies, highlighting key achievements and challenges, such as limited real-world validation, the need for hybrid approaches, and barriers to integrating ML into engineering workflows. By identifying critical research gaps and proposing future directions, this review aims to provide a comprehensive framework for advancing ML applications in structural engineering. The findings emphasize the transformative potential of ML to optimize design processes, enhance safety, and promote sustainable practices in civil engineering projects.

Predicting structural performance is a critical aspect of civil engineering, ensuring the safety, efficiency, and durability of buildings and infrastructure. Traditional methods, such as finite element analysis and empirical modeling, often fall short in addressing the complexities of modern structural systems. The advent of machine learning (ML) has revolutionized this domain by offering data-driven approaches capable of handling non-linear relationships and large datasets, enhancing the accuracy and efficiency of structural performance predictions. This review paper examines the applications of ML techniques, including Artificial Neural Networks (ANN), Support Vector Regression (SVR), Random Forest (RF), Decision Tree Regression (DTR), and hybrid models, in predicting structural metrics such as load-bearing capacity, deflection, durability, and seismic performance. The paper synthesizes findings from recent studies, highlighting key achievements and challenges, such as limited real-world validation, the need for hybrid approaches, and barriers to integrating ML into engineering workflows. By identifying critical research gaps and proposing future directions, this review aims to provide a comprehensive framework for advancing ML applications in structural engineering. The findings emphasize the transformative potential of ML to optimize design processes, enhance safety, and promote sustainable practices in civil engineering projects.

George Brocklehurst https://2017.northbaypython.org/schedule/presentation/20/ This talk will introduce a machine learning technique called linear regression, in a way that's targeted at working developers, not mathematicians or theoretical computer scientists. Want to get started with machine learning? This is how. Many of the principles of linear regression apply to other machine learning techniques, so the ground we cover today will set you up to explore the wider world of machine learning. About North Bay Python A single-track conference north of the Golden Gate, focused on community, collaboration, and all things Python. North Bay Python is a two-day, single-track Python conference held at the Mystic Theatre in Historic Downtown Petaluma, California, over the weekend of December 2 & 3, 2017. We're a nonprofit conference for professionals, enthusiasts and students alike. We're focused on inclusion, accessibility, diversity, and affordability. Most importantly, we're planning a great lineup of talks from all over the Python ecosystem, with plenty of time to meet new people and develop new ideas. Our venue, the Mystic Theatre in Downtown Petaluma, is a beautiful example of an early 1900s Vaudeville theatre. You can find over 50 different food and drink options a short walk away, and the nearest hotel is only a block away. A Python conference north of the Golden Gate North Bay Python is a single-track conference with a carefully curated set of talks representing the diverse Python community and their different areas of interest. If a topic is less to your interest, or you've met some people you really want to sit down and chat with, we'll have plenty of areas away from the main theatre to catch up and chat. Our goal is to keep prices as low as possible. That means we won't be catering lunch. Instead, you can look forward to extra-long lunch breaks you can use to explore all of the great food options around the venue.

The ability to navigate autonomously from the start to its final goal is the crucial key to mobile robots. To ensure complete navigation, it is mandatory to do heavy programming since this task is composed of several subtasks such as path planning, localization, and obstacle avoidance. This paper simplifies this heavy process by making the robot more intelligent. The robot will acquire the navigation policy from an expert in navigation using machine learning. We used the expert A*, which is characterized by generating an optimal trajectory. In the context of robotics, learning from demonstration (LFD) will allow robots, in general, to acquire new skills by imitating the behavior of an expert. The expert will navigate in different environments, and our robot will try to learn its navigation strategy by linking states and suitable actions taken. We find that our robot acquires the navigation policy given by A* very well. Several tests were simulated with environments of different complexity and obstacle distributions to evaluate the flexibility and efficiency of the proposed strategies. The experimental results demonstrate the reliability and effectiveness of the proposed method.

Machine Learning is part of Artificial Intelligence that has the ability to make future forecastings based on the previous experience. Methods has been proposed to construct models including machine learning algorithms such as Neural Networks (NN), Support Vector Machines (SVM) and Deep Learning. This paper presents a comparative performance of Machine Learning algorithms for cryptocurrency forecasting. Specifically, this paper concentrates on forecasting of time series data. SVM has several advantages over the other models in forecasting, and previous research revealed that SVM provides a result that is almost or close to actual result yet also improve the accuracy of the result itself. However, recent research has showed that due to small range of samples and data manipulation by inadequate evidence and professional analyzers, overall status and accuracy rate of the forecasting needs to be improved in further studies. Thus, advanced research on the accuracy rate of the forecasted price has to be done.

This paper introduces an intuitive approach to clickthrough rate (CTR) prediction, a learning problem that has been extensively studied over the past several years. As digital marketing continues to grow rapidly into a multi-billion-dollar industry, this study aims to find the most effective machine learning model to enhance the CTR of marketing emails by comparing various tree-based models. Key steps in this research include data collection, feature extraction, and CTR prediction through the evaluation of different models. The statistical results prove that the CatBoost model, with optimized feature selection, achieves near-perfect data fitting, indicating its efficiency.

This paper addresses the challenge of high computational complexity in video surveillance systems by proposing an efficient model that integrates hybrid machine learning algorithms (HML) for event recognition. Conventional surveillance methods struggle with processing vast amounts of video data in real-time, leading to scalability, and performance issues. Our proposed approach utilizes convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to enhance the accuracy and efficiency of detecting events. By comparing our model with conventional surveillance techniques motion detection, background subtraction, and frame differencing. We demonstrate significant improvements in frame processing time, object detection speed, energy efficiency, and anomaly detection accuracy. The integration of dynamic model scaling and edge computing further optimizes computational resource usage, making our method a scalable and effective solution for real time surveillance needs. This research highlights the potential of machine learning to revolutionize video surveillance, offering insights into developing more intelligent and responsive security systems. The results of your simulation analysis, indicating performance improvements in accuracy by 0.25%, 0.35%, and 0.45% for the motion detection algorithm, background subtraction, and frame differencing respectively, and in real-time data processing by 5.65%, 4.45%, and 6.75% for the motion detection algorithm, background subtraction, and frame differencing respectively, highlight the potential of machine learning to transform video surveillance into a more intelligent and responsive system.

MLOps Coffee Sessions #96 with Sebastián Ramírez, FastAPI for Machine Learning co-hosted by Adam Sroka. // Abstract Fast API almost never happened. Sebastián Ramírez, the creator of FastAPI, tried as hard as possible not to build something new. After many failed attempts at finding what he was looking for he decided to scratch his own itch and build a new product.    The conversation goes over what Fast API is, how Sebastián built it, what the next big problems to tackle in ML are, and how to focus on adding value where you can. // Bio ?? Sebastián Ramírez is the creator of FastAPI, Typer, and other open-source tools. Currently, Sebastián is a Staff Software Engineer at Forethought while also helping other companies as an external consultant.?? // MLOps Jobs board   https://mlops.pallet.xyz/jobs // Related Links Website: https://tiangolo.com/ https://fastapi.tiangolo.com/ https://typer.tiangolo.com/ https://www.forethought.ai/ https://sqlmodel.tiangolo.com/ https://github.com/tiangolo --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Adam on LinkedIn: https://www.linkedin.com/in/aesroka/ Connect with Sebastián on LinkedIn: https://www.linkedin.com/in/tiangolo/ Timestamps: [00:00] Introduction to Sebastián Ramírez [00:44] Takeaways [02:45] Apply () Conference is coming up! [03:38] FastAPI background [05:02] Ramp up reason [06:17] Tipping point [08:11] Surprising ways using FastAPI [10:08] Twist it and break it lessons learned [12:00] Length of comprehension process [15:59] Missing pieces [21:25] Advice to technically capable what to start with [25:19] Making FastAPI better [27:52] What to simplify and why are they cumbersome right now? [30:14] Building FastAPI vs solving the problem [32:42] Next itch to scratch [34:26] Landscape's pathway [38:03] Things that would not change [40:13] Sebastián's change in life since FastAPI [43:09] Sebastián's famous tweet [44:13] Experienced vs inexperienced [46:07] Approach to becoming a tools expert [50:22] Wrap up

Game AI developers tend to treat machine learning with a mixture of curiosity, skepticism and dismissal. There's a widespread perception that certain unique aspects of game AI make it a poor fit for ML. However, game AI is far from unique in its needs. The generally bad experiences game AI has had with ML are primarily due to a lack of knowledge about the necessary tools and techniques rather than a failing of the technique itself. Simply understanding these tools and techniques, however, often helps game devs get beyond the suspicion and reluctance and use machine learning techniques to improve their games. This lecture will describe key concepts from ML, such as over-fitting and the bias/variance tradeoff, and the problem spaces (supervised/unsupervised, classification versus regression, etc.). We will then give a quick overview of several common and valuable ML algorithms.

The "2nd Annual WPI-UMASS-UPENN EDM Data Mining Challenge" required contestants to predict efficient testtaking based on log data. In this paper, we describe our theory-driven and psychometric modeling approach. For feature engineering, we employed the Log-Normal Response Time Model for estimating latent person speed, and the Generalized Partial Credit Model for estimating latent person ability. Additionally, we adopted an n-gram feature approach for event sequences. For training a multi-label classifier, we distinguished inefficient test takers who were going too fast and those who were going too slow, instead of using the provided binary target label. Our best-performing ensemble classifier comprised three sets of low-dimensional classifiers, dominated by test-taker speed. While our classifier reached moderate performance, relative to competition leaderboard, our approach makes two important contributions. First, we show how explainable classifiers could provide meaningful predictions if results can be contextualized to test administrators who wish to intervene or take action. Second, our re-engineering of test scores enabled us to incorporate person ability into the estimation. However, ability was hardly predictive of efficient behavior, leading to the conclusion that the target label's validity needs to be questioned. The paper concludes with tools that are helpful for substantively meaningful log data mining. [For the full proceedings, see ED607784.]

Neural networks are ubiquitous in applied machine learning for education. Their pervasive success in predictive performance comes alongside a severe weakness, the lack of explainability of their decisions, especially relevant in humancentric fields. We implement five state-of-the-art methodologies for explaining black-box machine learning models (LIME, PermutationSHAP, KernelSHAP, DiCE, CEM) and examine the strengths of each approach on the downstream task of student performance prediction for five massive open online courses. Our experiments demonstrate that the families of explainers do not agree with each other on feature importance for the same Bidirectional LSTM models with the same representative set of students. We use Principal Component Analysis, Jensen-Shannon distance, and Spearman's rank-order correlation to quantitatively cross-examine explanations across methods and courses. Furthermore, we validate explainer performance across curriculum-based prerequisite relationships. Our results come to the concerning conclusion that the choice of explainer is an important decision and is in fact paramount to the interpretation of the predictive results, even more so than the course the model is trained on. Source code and models are released at http://github.com/epfl-ml4ed/evaluating-explainers. [For the full proceedings, see ED623995.]

Elpasolite is the predominant quaternary crystal structure (AlNaK$_2$F$_6$ prototype) reported in the Inorganic Crystal Structure Database. We have developed a machine learning model to calculate density functional theory quality formation energies of all $\sim$2 M pristine ABC$_2$D$_6$ elpasolite crystals which can be made up from main-group elements (up to bismuth). Our model's accuracy can be improved systematically, reaching 0.1 eV/atom for a training set consisting of 10 k crystals. Important bonding trends are revealed, fluoride is best suited to fit the coordination of the D site which lowers the formation energy whereas the opposite is found for carbon. The bonding contribution of elements A and B is very small on average. Low formation energies result from A and B being late elements from group (II), C being a late (I) element, and D being fluoride. Out of 2 M crystals, 90 unique structures are predicted to be on the convex hull---among which NFAl$_2$Ca$_6$, with peculiar stoichiometry and a negative atomic oxidation state for Al.

Privacy-preserving distributed machine learning becomes increasingly important due to the recent rapid growth of data. This paper focuses on a class of regularized empirical risk minimization (ERM) machine learning problems, and develops two methods to provide differential privacy to distributed learning algorithms over a network. We first decentralize the learning algorithm using the alternating direction method of multipliers (ADMM), and propose the methods of dual variable perturbation and primal variable perturbation to provide dynamic differential privacy. The two mechanisms lead to algorithms that can provide privacy guarantees under mild conditions of the convexity and differentiability of the loss function and the regularizer. We study the performance of the algorithms, and show that the dual variable perturbation outperforms its primal counterpart. To design an optimal privacy mechanisms, we analyze the fundamental tradeoff between privacy and accuracy, and provide guidelines to choose privacy parameters. Numerical experiments using customer information database are performed to corroborate the results on privacy and utility tradeoffs and design.

Poor (even random) starting points for learning/training/optimization are common in machine learning. In many settings, the method of Robbins and Monro (online stochastic gradient descent) is known to be optimal for good starting points, but may not be optimal for poor starting points -- indeed, for poor starting points Nesterov acceleration can help during the initial iterations, even though Nesterov methods not designed for stochastic approximation could hurt during later iterations. The common practice of training with nontrivial minibatches enhances the advantage of Nesterov acceleration.