OMRON HeartGuide
January 1, 2022Article of the Month – January 2022
January 1, 2022Microsoft A.I
Robust framework for developing innovative solutions in machine learning and data sciences. They support researchers using AI to improve the world...
Overview
Microsoft AI offers innovative solutions in machine learning data operation, sciences, robotics, and other technologies. They also focus on supporting organizations that apply AI in their platforms to improve their contribution to multiple disciplines, including the environment, accessibility, humanitarian problems, and healthcare issues. They are committed to advancing AI by providing tools, grants, and expertise to inspire organizations working to solve current problems. Their objective is to help individuals and companies turn ideas into tangible results
Microsoft AI Principles
The Office of Responsible AI (ORA) and the AI, Ethics, and Effects in Engineering and Research (Aether) are the committees that use Microsoft’s AI principles in their business. These Committees counsel their users on the challenges and benefits presented by AI technologies. ORA establishes the frame rules in all processes, working closely with all the company’s components to achieve the goal.
The result is that teams and individuals work closely to advocate and uphold Microsoft AI principles in their daily activities.
- Fairness: AI systems should treat people equally.
- Reliability and safety: AI systems should perform reliably and safely.
- Privacy and Security: AI systems should guaranty privacy and security
The Office of Responsible AI (ORA) and the AI, Ethics, and Effects in Engineering and Research (Aether) are the committees that use Microsoft’s AI principles in their business. These Committees counsel their users on the challenges and benefits presented by AI technologies. ORA establishes the frame rules in all processes, working closely with all the company’s components to achieve the goal.
The result is that teams and individuals work closely to advocate and uphold Microsoft AI principles in their daily activities.
- Fairness: AI systems should treat people equally.
- Reliability and safety: AI systems should perform reliably and safely.
- Privacy and Security: AI systems should guaranty privacy and security
- Inclusiveness: AI systems should facilitate people’s lives and incentivize engagement
- Transparency: AI systems should be easy to understand.
- Accountability: developers of AI systems should be held accountable.
ORA applies these responsibility principles and puts them to practice by establishing a series of company rules and implementing a governance system for both their private and public businesses. It has four key functions:
- Governance: establishing clear rules for enacting responsible AI and defining specific roles and responsibilities for everyone involved.
- Team enablement: predisposition to adopt responsible practices, within the company, with customers, coworkers, and associates.
- Sensitive use cases: reviewing the use of AI in specific sensitive cases ensures that Microsoft AI principles are maintained during the developing process.
4. Public policy: to ensure the ethical application of AI technology, the creation of laws, norms, principles, and standards is a priority for the benefit of everyone (1)
Microsoft IT Aspects
Data and AI: Microsoft CSEO has created a platform to acquire and visualize spatial data at all Microsoft facilities to improve data-driven decisions around allocating physical space. Achievements with machine learning include data-driven decisions, meaningful conversations, efficiency, and productivity. The general approach includes retrieving information from multiple sources while using existing infrastructure: badges, Wi-Fi telemetry, employee tools, and weather data.
Microsoft 365: Consists of a platform with numerous products. It has a systematic operating system, security tools, and productivity apps.
Azure and Cloud infrastructure: More than 200 products and cloud services form the Azure cloud platform. It can also provide an environment with tools and services designed for this hybrid cloud and build, debug, deploy, and manage applications with the language or platform of the customers’ choice.
Data and AI: Microsoft CSEO has created a platform to acquire and visualize spatial data at all Microsoft facilities to improve data-driven decisions around allocating physical space. Achievements with machine learning include data-driven decisions, meaningful conversations, efficiency, and productivity. The general approach includes retrieving information from multiple sources while using existing infrastructure: badges, Wi-Fi telemetry, employee tools, and weather data.
Microsoft 365: Consists of a platform with numerous products. It has a systematic operating system, security tools, and productivity apps.
Azure and Cloud infrastructure: More than 200 products and cloud services form the Azure cloud platform. It can also provide an environment with tools and services designed for this hybrid cloud and build, debug, deploy, and manage applications with the language or platform of the customers’ choice.
It also contains machine learning models and tools for various skill levels, such as visual drag and drop and zero-code automated models, to full-model development and algorithm selection.
Business applications: Applications like Dynamics 365, Power Automate, Power BI, and Power Virtual Agents allow users to develop ideas in short periods.
Security: Zero Trust architecture directed the modern security challenges included with cloud migration and a mobile workforce. By using Zero Trust, Microsoft takes a layered approach to secure corporate and customer data. Zero Trust includes user identity security, device health verification, validation of application, and secure access of the least-privileged to corporate resources and services.(2)
Microsoft as a Technology Company
Last year Microsoft’s revenue was 143 billion, $53 billion in operating income, and more than $60 billion in operating cash flow. The return for investors was more than $30 billion.
Their commercial cloud service raised $50 million in revenue, an increase of 36% compared to the previous year.
Microsoft bought GitHub in 2018 and is planning to invest in building the most complete toolchain for developers, independent of language, framework, or cloud.
For 2019, 95% of the Fortune 500 trust Azure for the mission-critical workloads. Microsoft is one of the first companies to open cloud data centers in South Africa and the Middle East.
As a technology company, Microsoft has affirmed to operate carbon neutral across worldwide operations every year since 2012. Their objective is to reduce operational emissions by 75% by 2030.(3)
Last year Microsoft’s revenue was 143 billion, $53 billion in operating income, and more than $60 billion in operating cash flow. The return for investors was more than $30 billion. Their commercial cloud service raised $50 million in revenue, an increase of 36% compared to the previous year.
Microsoft bought GitHub in 2018 and is planning to invest in building the most complete toolchain for developers, independent of language, framework, or cloud.
For 2019, 95% of the Fortune 500 trust Azure for the mission-critical workloads. Microsoft is one of the first companies to open cloud data centers in South Africa and the Middle East.
As a technology company, Microsoft has affirmed to operate carbon neutral across worldwide operations every year since 2012. Their objective is to reduce operational emissions by 75% by 2030.(3)
Machine Learning that can recognize Patterns of Engagement in Mental Health
Machine Learning was used to provide a graphic framework for different behaviors under cognitive-behavioral therapy intervention (iCBT). Consequently, a study using data from 54,604 individuals identified five heterogeneous subtypes associated with different patterns of patient behavior and different levels of improvement in symptoms of depression and anxiety.(4)
Using AI for COVID-19 Chest X-Ray Diagnosis
Using a Machine Learning system called Microsoft CustomVision, a research group developed a model capable of diagnosing COVID-19 cases.
The model was trained using 103 CRX images of COVID-19 pneumonia, 500 images of non- COVID-19 pneumonia, and 500 images of normal lungs. To validate the dataset, 30 random images were obtained. The dataset was performed on ten images from hospitalized COVID-19 patients, the CXR pneumonia images from non-COVID-19, and ten normal CXRs. After training, the model performed with a sensitivity of 100%, a specificity of 95%, an accuracy of 97%, a 91% positive predictive value (PPV), and a 100% negative predictive value.
Using AI for COVID-19 Chest X-Ray Diagnosis
Using a Machine Learning system called Microsoft CustomVision, a research group developed a model capable of diagnosing COVID-19 cases.
The model was trained using 103 CRX images of COVID-19 pneumonia, 500 images of non- COVID-19 pneumonia, and 500 images of normal lungs. To validate the dataset, 30 random images were obtained. The dataset was performed on ten images from hospitalized COVID-19 patients, the CXR pneumonia images from non-COVID-19, and ten normal CXRs. After training, the model performed with a sensitivity of 100%, a specificity of 95%, an accuracy of 97%, a 91% positive predictive value (PPV), and a 100% negative predictive value.
This technology has applications for screening, triage, initial diagnosis, monitoring, and risk identification, showing the potential for how AI could change the practice and application of medicine in the future. (5)
Advertising Systems Can Identify People Suffering from Serious Medical Conditions
The internet ads systems could learn to screen for lung, breast, and colon cancer and ultimately could find people who are likely to have cancer. Thus, clinically verified questionnaires can be used to calculate the potential of having a suspected cancer diagnosis.(6)
A Machine Learning Algorithm Successfully Screens for Parkinson’s in Web Users
A supervised ML system learned to distinguish internet users with self-reported Parkinson’s disease from controls based on their interaction (including contest of the search terms, locations of the mouse pointer over time, mouse clicks, and interactions with the keyboard) with a search engine called Bing. For ethical reasons, users and researchers did not have access to any database lignin anonymized user identifiers with IP addresses. As a screening tool, the model had a PPV of 25%, resulting in 17,843 of 1,490,987 web users over the age of 40 years screening positive for Parkinson’s disease. This percentage was higher in at-risk groups.
These results indicate that an automatic classifier, based on mouse and keyboard interactions, can reliably trace individuals at high risk for Parkinson’s disease and identify the progression of disease-related signs in those screened positive. (7)
Predicting Length of Stay in Intensive Care Unit Using AI
Efficient allocation of resource-heavy Intensive Care Unit (UCI) beds is critical in the increasingly demanding and budget restricting health institutions. A central component to solve this problem is knowing how long the current patients in the ICU are likely to stay.
A deep learning model called Temporal Pointwise Convolution was designed with this objective. Data was collected from 208 hospitals across the United States, and all patients were adults (>18 years), with a length of stay of at least 5 hours and at least one recorded observation. Other data collected included diagnoses, gender, age, the hour of admission, and ethnicity. The system achieved significant performance benefits of 18-51% compared to other predicting models. (8)
Predicting Length of Stay in Intensive Care Unit Using AI
Efficient allocation of resource-heavy Intensive Care Unit (UCI) beds is critical in the increasingly demanding and budget restricting health institutions. A central component to solve this problem is knowing how long the current patients in the ICU are likely to stay.
A deep learning model called Temporal Pointwise Convolution was designed with this objective. Data was collected from 208 hospitals across the United States, and all patients were adults (>18 years), with a length of stay of at least 5 hours and at least one recorded observation. Other data collected included diagnoses, gender, age, the hour of admission, and ethnicity. The system achieved significant performance benefits of 18-51% compared to other predicting models. (8)
Understanding the Dynamics of Infectious Disease Spread in a Heterogeneous Population
A multi-compartment version of the classic Susceptible Infected–Recovered (SIR) model was designed using data on Respiratory Syncytial Virus and West Nile Virus to predict the spread of infectious diseases. By capturing the spatial and temporal dynamics of these viruses, the pathogen’s spread could be predicted.(9)