Todays organisations are witnessing an accelerated pace in building Machine Learning-fuelled solutions, so much so that ML has quickly become the most acquired skill in India on online learning platforms. An implicit requirement for the teams involved in an organisation is working collaboratively while constantly building and managing a large number of models.

To that end, Microsoft India hosted a webinar titled Machine Learning lifecycle management and possibilities with Azure Machine Learning, featuring Aruna Chakkirala, Senior Cloud Solutions Architect, Microsoft India, in order to show the benefits of Azure Machine Learning and how it enables data science and IT teams to collaborate and increase the pace of model development and deployment via monitoring, validation, and governance of machine learning models.

The webinar came with some amazing insights about Azure Machine Learning (AML) and its properties, along with a quick demo.

Heres what the webinar covered:

The idea is to make machine learning available to data scientists and developers of all skill levels, as well as provide an end-to-end lifecycle management for machine learning through MLOps. It also enables responsible and trustworthy development of machine learning solutions through responsible AI/ML capabilities which provides transparency and explainability for the models, said Aruna.

Azure Machine Learning is comprehensive, giving users end-to-end capabilities, extensive ML platform abilities in Azure, which empowers data scientists and developers with a wide range of predictive experiences for building, training, and deploying ML models both securely and at scale.

Aruna next explained the ML lifecycle. When it comes to creating ML platforms, the process begins with data scientists and data engineers working in tandem, figuring out what part of the data is interesting. The data scientist will then take the insights to build a model from it. That is followed by the registry of the model. And finally, users can go ahead and release this model into production. The model will need to be monitored periodically. Because once the model is monitored, youll get insights about how it fares in the real world scenario.

AML is a set of cloud services combined with an AML studio interface and also an SDK which brings together the power of what you need to build services and take them into production. It enables users to prepare the data, build, train, manage, track, and deploy models, Aruna said.

The AutoML option in AML studio is a quick and easy way to build models, primarily aimed at the citizen data scientist. All it requires is a dataset, and using the wizard to set up a few configurations, AutoML does all the work in building multiple models, identifying the best model based on the metrics and providing a view of the multiple runs. All this is possible in just a few clicks.

AML Designer is an UI interface that enables users to build machine learning pipelines with drag-n-drop experience and simplify the publishing and deployment of pipelines.

It helps users connect to their own data with ease, hundreds of pre-built components help build and train models without writing code, it helps automate model validation, evaluation and interpretation in their pipeline and enables deployment of models and publishment of endpoints with a few clicks.

Azure ML Service workspace comes with a lot of components such as models, experiments, pipelines, compute target, environments, deployment, datastores and data labeling.

Aruna then conducted a quick demo, showing the components of AML. The ML service has a number of components within it, such as values, resource group, storage, studio web url, registry, application insights and ML flow.

Coming to AML studio, it too has various core components. It has the option to provide compute, where users are creating the training resources. The studio also comes with curated environments, data store, date labeling, and link services. This demo was followed by a quick guided demo of AutoML as well.

Aruna added that AML can implement an end-to-end ML lifecycle. It ties down the entire ML lifecycle for us. Every stage of the ML lifecycle has been tied together within the whole spectrum of components, she said.

Workflow steps:

1. Develop machine learning training scripts in Python, using autoML, Designer, Notebooks, etc.

2. Create and configure a compute target.

3. Submit the scripts to the configured compute target to run in that environment. During training, the compute target stores run records to a datastore. There, the records are saved to an experiment.

4. Review the experiment for logged metrics from the current and past runs.

5. Once a satisfactory run is found, register the persisted model in the registry.

6. Develop a scoring script.

7. Create an image and register it in the image registry.

8. Deploy the image as a web service in Azure.

9. Monitor the model in production and identify when further improvements are required.

There are three ways to deploy your model in Azure ML - Through real-time interference such as HTTP endpoints, batch interference, and through managed endpoints.

Monitoring on AML has 23 metrics that you can choose from, across various categories like model, resource, run and quota. Users can look at monitoring from two different perspectives:

1. Monitoring as an administrator, which means monitoring the health of the resource, monitoring compute quota, etc.

2. Monitoring as a data scientist or developer, which means monitoring training runs, tracking experiments and visualising runs.

You don't want to be repeating the same steps again and thats where MLOps comes into play. It removes the drudgery of repeated processes, said Aruna. It brings DevOps principles into the ML world. It brings together people, processes, and platforms to automate ML-infused software delivery and provide continuous value to users.

MLOps comes with multiple benefits such as automation and observability, validation and reproducibility.

Aruna noted that model interpretability and fairness are the cornerstones of Azure Machine Learnings AI/ML offerings. As machine learning becomes ubiquitous in decision making, it becomes extremely necessary to provide tools which can bring out model transparency, she said.

Read the original:
Azure Machine Learning Webinar recap: A sneak peek at the process and possibilities of AML - YourStory

Google explains how it keeps user-created reviews on Google Maps free of fraud and abuse in a new blog post and accompanying video. Like many platforms dealing with moderation at scale, Google says it uses a mix of automated machine learning systems as well as human operators.

The details come amidst growing scrutiny of user reviews on sites like Google Maps and Yelp, where businesses have been hit with bad reviews for implementing COVID-related health and safety measures (including mask and vaccine requirements) often beyond their control. Other reviews have criticized businesses for supposedly leading them to contract COVID-19 or for not keeping to usual business hours during a global pandemic.

Earlier today, Yelp reported that it removed over 15,500 reviews between April and December last year for violating its COVID-19 content guidelines, a 161 percent increase over the same period in 2020. In total, Yelp says it removed over 70,200 reviews across nearly 1,300 pages in 2021, with many resulting from so-called review bombing incidents where coordinated reviews are submitted from users who havent actually patronized a business.

Google explains that every review posted on Google Maps is checked by its machine learning system, which has been trained on the companys content policies to weed out abusive or misleading reviews. This system is trained to check both the contents of individual reviews, but itll also look for wider patterns like sudden spikes in one- or five-star reviews both from the account itself, as well as other reviews on the business.

Google says that human moderation comes into play for content thats been flagged by end users and businesses themselves. Offending reviews can be removed, and in more severe cases, user accounts can be suspended and litigation pursued. Weve found that we need both the nuanced understanding that humans offer and the scale that machines provide to help us moderate contributed content, Googles product lead for user-generated content, Ian Leader, writes.

Its an interesting look at the steps Google takes to keep Maps reviews usable. You can read more in the full blog post.

Link:
Google Maps review moderation detailed as Yelp reports thousands of violations - The Verge

Scientists at Rice University are using machine-learning techniques to fine-tune the process of synthesizing graphene from waste through flash Joule heating. The researchers describe in their new work how machine-learning models that adapt to variables and show them how to optimize procedures are helping them push the technique forward.

Machine learning is fine-tuning Rice Universitys flash Joule heating method for making graphene from a variety of carbon sources, including waste materials. Credit: Jacob Beckham, from: Phys.org

The process, discovered by the Rice lab of chemist James Tour, has expanded beyond making graphene from various carbon sources to extracting other materials like metals from urban waste, with the promise of more environmentally friendly recycling to come. The technique is the same: blasting a jolt of high energy through the source material to eliminate all but the desired product. However, the details for flashing each feedstock are different.

"Machine-learning algorithms will be critical to making the flash process rapid and scalable without negatively affecting the graphene product's properties," Prof. Tour said.

"In the coming years, the flash parameters can vary depending on the feedstock, whether it's petroleum-based, coal, plastic, household waste or anything else," he said. "Depending on the type of graphene we wantsmall flake, large flake, high turbostratic, level of puritythe machine can discern by itself what parameters to change."

Because flashing makes graphene in hundreds of milliseconds, it's difficult to follow the details of the chemical process. So Tour and company took a clue from materials scientists who have worked machine learning into their everyday process of discovery.

"It turned out that machine learning and flash Joule heating had really good synergy," said Rice graduate student and lead author Jacob Beckham. "Flash Joule heating is a really powerful technique, but it's difficult to control some of the variables involved, like the rate of current discharge during a reaction. And that's where machine learning can really shine. It's a great tool for finding relationships between multiple variables, even when it's impossible to do a complete search of the parameter space". "That synergy made it possible to synthesize graphene from scrap material based entirely on the models' understanding of the Joule heating process," he explained. "All we had to do was carry out the reactionwhich can eventually be automated."

The lab used its custom optimization model to improve graphene crystallization from four starting materialscarbon black, plastic pyrolysis ash, pyrolyzed rubber tires and cokeover 173 trials, using Raman spectroscopy to characterize the starting materials and graphene products.

The researchers then fed more than 20,000 spectroscopy results to the model and asked it to predict which starting materials would provide the best yield of graphene. The model also took the effects of charge density, sample mass and material type into account in their calculations.

Lat month, the Rice team developed an acoustic processing method to analyze LIG synthesis in real time.

Read more:
Machine learning helps improve the flash graphene process - Graphene-Info

Summary: A new deep learning algorithm that looks for the burden of genomic variants is 70% accurate at identifying specific mental health disorders within the African-American community.

Source: CHOP

Minority populations have been historically under-represented in existing studies addressing how genetic variations may contribute to a variety of disorders. A new study from researchers at Childrens Hospital of Philadelphia (CHOP) shows that a deep learning model has promising accuracy when helping to diagnose a variety of common mental health disorders in African American patients.

This tool could help distinguish between disorders as well as identify multiple disorders, fostering early intervention with better precision and allowing patients to receive a more personalized approach to their condition.

The study was recently published by the journalMolecular Psychiatry.

Properly diagnosing mental disorders can be challenging, especially for young toddlers who are unable to complete questionnaires or rating scales. This challenge has been particularly acute in understudied minority populations. Past genomic research has found several genomic signals for a variety of mental disorders, with some serving as potential therapeutic drug targets.

Deep learning algorithms have also been used to successfully diagnose complex diseases like attention deficit hyperactivity disorder (ADHD). However, these tools have rarely been applied in large populations of African American patients.

In a unique study, the researchers generated whole genome sequencing data from 4,179 patient blood samples of African American patients, including 1,384 patients who had been diagnosed with at least one mental disorder This study focused on eight common mental disorders, including ADHD, depression, anxiety, autism spectrum disorder, intellectual disabilities, speech/language disorder, delays in developments and oppositional defiant disorder (ODD).

The long-term goal of this work is to learn more about specific risks for developing certain diseases in African American populations and how to potentially improve health outcomes by focusing on more personalized approaches to treatment.

Most studies focus only on one disease, and minority populations have been very under-represented in existing studies that utilize machine learning to study mental disorders, said senior author Hakon Hakonarson, MD, Ph.D., Director of the Center for Applied Genomics at CHOP.

We wanted to test this deep learning model in an African American population to see whether it could accurately differentiate mental disorder patients from healthy controls, and whether we could correctly label the types of disorders, especially in patients with multiple disorders.

The deep learning algorithm looked for the burden of genomic variants in coding and non-coding regions of the genome. The model demonstrated over 70% accuracy in distinguishing patients with mental disorders from the control group. The deep learning algorithm was equally effective in diagnosing patients with multiple disorders, with the model providing exact diagnostic matches in approximately 10% of cases.

The model also successfully identified multiple genomic regions that were highly enriched formental disorders, meaning they were more likely to be involved in the development of these medical disorders. The biological pathways involved included ones associated with immune responses, antigen and nucleic acid binding, a chemokine signaling pathway, and guanine nucleotide-binding protein receptors.

However, the researchers also found that variants in regions that did not code for proteins seemed to be implicated in these disorders at higher frequency, which means they may serve as alternative markers.

By identifying genetic variants and associated pathways, future research aimed at characterizing their function may provide mechanistic insight as to how these disorders develop, Hakonarson said.

Author: Press OfficeSource: CHOPContact: Press Office CHOPImage: The image is in the public domain

Original Research: Open access.Application of deep learning algorithm on whole genome sequencing data uncovers structural variants associated with multiple mental disorders in African American patients by Yichuan Liu et al. Molecular Psychiatry

Abstract

Application of deep learning algorithm on whole genome sequencing data uncovers structural variants associated with multiple mental disorders in African American patients

Mental disorders present a global health concern, while the diagnosis of mental disorders can be challenging. The diagnosis is even harder for patients who have more than one type of mental disorder, especially for young toddlers who are not able to complete questionnaires or standardized rating scales for diagnosis. In the past decade, multiple genomic association signals have been reported for mental disorders, some of which present attractive drug targets.

Concurrently, machine learning algorithms, especially deep learning algorithms, have been successful in the diagnosis and/or labeling of complex diseases, such as attention deficit hyperactivity disorder (ADHD) or cancer. In this study, we focused on eight common mental disorders, including ADHD, depression, anxiety, autism, intellectual disabilities, speech/language disorder, delays in developments, and oppositional defiant disorder in the ethnic minority of African Americans.

Blood-derived whole genome sequencing data from 4179 individuals were generated, including 1384 patients with the diagnosis of at least one mental disorder. The burden of genomic variants in coding/non-coding regions was applied as feature vectors in the deep learning algorithm. Our model showed ~65% accuracy in differentiating patients from controls. Ability to label patients with multiple disorders was similarly successful, with a hamming loss score less than 0.3, while exact diagnostic matches are around 10%. Genes in genomic regions with the highest weights showed enrichment of biological pathways involved in immune responses, antigen/nucleic acid binding, chemokine signaling pathway, and G-protein receptor activities.

A noticeable fact is that variants in non-coding regions (e.g., ncRNA, intronic, and intergenic) performed equally well as variants in coding regions; however, unlike coding region variants, variants in non-coding regions do not express genomic hotspots whereas they carry much more narrow standard deviations, indicating they probably serve as alternative markers.

See the original post:
Using Deep Learning to Find Genetic Causes of Mental Health Disorders in an Understudied Population - Neuroscience News