Untitled video - Made with Clipchamp

Introduction

Professor from Purdue University introduced in a seminar focusing on machine learning.
Received PhD from Brown University in 2007 and worked at DOE in various leadership roles. Currently serves as associate dean for science and engineering research.

Overview of Machine Learning Work

The professor's work in machine learning involves outputting two channels instead of the traditional single channel:
- One channel provides the main estimation.
- The second channel provides the confidence interval, indicating the uncertainty in predictions.
Emphasizes feedback based on the confidence interval to enhance model performance and predict reliability.

Waves of AI Defined by DARPA

First Wave (1990s)

Early AI applications like chess-playing programs (deep blue).

Second Wave (2000s-Present)

Introduced sophisticated systems like AlphaGo, capable of playing Go against champions.

Third Wave (Present)

Focus on large language models (e.g., GPT); continued improvement in learning and reasoning.
Noted that reasoning is still an area that requires further development.

Fourth Wave (2030 and beyond)

Aim for artificial general intelligence (AGI), enabling AI to perform any cognitive task a human can do.

Applications of AI

AI's ability extends to tasks like:
- Coloring old photographs.
- Autonomous driving (as exemplified by Tesla).
Discussed projects applying AI for social good, including:
- Health digital twins targeting pediatric cardiovascular diseases.
- Energy management in smart grids accommodating renewables like wind and solar.

Research on AI Robustness

Work funded by the Simon Foundation aimed to ensure AI systems can operate safely in real-world scenarios, like addressing challenges in autonomous driving where external signals might be manipulated (e.g., hackers altering traffic signs).

AI in Biomedical Applications

Discussed a project using AI in conjunction with ultrasound to automate heart boundary detection:
- Aims for real-time, high-resolution imaging, enhancing speed and cost-effectiveness.
Collaboration with companies like Philips and Fujifilm aims to commercialize these techniques.

Data Science Education Ecosystem at Purdue

Initiatives to foster data mining communities among undergraduates:
- Creating learning environments to encourage hands-on collaboration and interaction with industry mentors,
- Structured dormitory settings to cultivate learning.

Historical Perspectives on AI and ML

AI Development and Models

Overview of key milestones in AI:
- 1955: Early AI programs using reasoning.
- 1990s Deep Blue.
- 2012 and beyond: Breakthroughs with AlexNet, ResNet, AlphaFold, and continued innovations.

Issues with Deep Learning

Challenges with non-convex optimization and local minima in training deep learning models.
Emphasized the importance of effective training methods and the need for graduate student support in AI projects.

Future Directions in AI

Integral AI

Aimed at deriving scientific knowledge from noisy data, with examples relevant to COVID-19 models.
Discussed unique epidemical modeling approaches that incorporate vaccination impacts and decision-making support for public health.

AI's Role in Alzheimer’s Research

Development of causal models rooted in biomarkers to predict patient deterioration.
Efforts to analyze individual parameters for personalized treatment models, enhancing diagnostic capabilities.

Conclusion

Reinforced the collaborative efforts at Purdue to build a landscape for AI applications in healthcare and engineering.
Mentioned the drive to contribute positively to society through research and education.

Introduction

Professor from Purdue University introduced in a seminar focusing on advanced machine learning methodologies and their real-world applications. The professor received a PhD from Brown University in 2007, followed by extensive work at the Department of Energy (DOE) in various significant leadership roles. Currently, he serves as the associate dean for science and engineering research, where he oversees the development of interdisciplinary research programs that bridge the gap between technology and societal needs.

Overview of Machine Learning Work

The professor's work in machine learning is innovative and cutting-edge, particularly in the development of models that output two channels instead of the traditional single channel used in many machine learning applications. This dual-channel output includes:

Primary Channel: Provides the main estimation outcomes, which represent the best guess or prediction based on the input data.
Confidence Interval Channel: Offers a measure of uncertainty around the predictions, indicating the reliability of the model's output and the likelihood of various potential outcomes. This dual output system is designed to enhance the robustness and interpretability of machine learning models, allowing for more informed decision-making. The professor emphasizes the importance of incorporating feedback mechanisms based on the confidence interval, which can significantly improve model performance and trust in predictions.

Waves of AI Defined by DARPA

First Wave (1990s): This phase marks the advent of early AI applications characterized by foundational technology, such as chess-playing programs like IBM's Deep Blue, which showcased the potential of AI in strategic thinking and decision-making.
Second Wave (2000s-Present): This era introduced more sophisticated AI systems, epitomized by AlphaGo, which could play Go, a complex board game, at a championship level, defeating world champions and demonstrating advanced learning capabilities.
Third Wave (Present): Focus is shifting towards large language models (LLMs) such as GPT, which have enhanced abilities in natural language processing, learning, and reasoning. Despite these advancements, reasoning skills remain an area that requires further development and refinement.
Fourth Wave (2030 and beyond): The goal of this wave is to achieve artificial general intelligence (AGI), the capacity for AI to perform any cognitive task that a human can undertake, marking a significant milestone in AI evolution.

Applications of AI

AI applications are diverse and impactful, extending to various innovative tasks, including:

Coloring Old Photographs: Utilizing advanced algorithms to restore and enhance historical photos with realistic colors.
Autonomous Driving: Technologies developed for vehicles, with Tesla being a prominent example, showcasing the potential of AI in transportation.
Projects for Social Good: Notable initiatives discussed include:
- Health digital twins designed to target pediatric cardiovascular diseases, creating predictive models for patient care.
- Energy management systems in smart grids that facilitate the integration of renewable energy sources like wind and solar, optimizing energy distribution and consumption.

Research on AI Robustness

The professor oversees work funded by the Simon Foundation, focusing on the robustness of AI systems to ensure that they can maintain performance under various real-world conditions. This includes addressing vulnerabilities in autonomous driving systems where external signals could be manipulated (like hackers altering traffic signs), highlighting the importance of security and reliability in AI applications.

AI in Biomedical Applications

A significant project involves the use of AI in conjunction with ultrasound technology to automate heart boundary detection. This project aims to achieve:

Real-Time, High-Resolution Imaging: Enhancing the speed and accuracy of medical imaging processes.
Collaboration with Industry Leaders: Working with companies such as Philips and Fujifilm to bring these advanced imaging techniques to market, affording healthcare professionals better tools for patient diagnosis.

Data Science Education Ecosystem at Purdue

Purdue University fosters a supportive environment for data science education, with initiatives aimed at building data mining communities among undergraduates. Key elements include:

Learning contexts that nurture hands-on collaboration among students and facilitate interaction with industry mentors.
Structured living and learning dormitories designed to cultivate an immersive educational experience focused on data science and machine learning.

Historical Perspectives on AI and ML

AI Development and Models

An overview of key milestones in the evolution of artificial intelligence includes:

1955: Emergence of early AI programs that began exploring reasoning.
1990s: The development of Deep Blue and other pioneering AI technologies.
2012 and Beyond: Breakthrough innovations such as AlexNet, ResNet, AlphaFold, and continuous advancements that push the boundaries of what AI can achieve.

Issues with Deep Learning

The professor emphasizes challenges associated with non-convex optimization in deep learning, particularly the existence of local minima that may impede the training process of deep learning models. This reinforces the call for effective training methods and the necessity of graduate student involvement in AI research projects to drive innovation.

Future Directions in AI

Integral AI

Integral AI seeks to extract meaningful scientific knowledge from noisy data, with particular relevance to pandemic-related models such as those centered around COVID-19. This includes:

Unique approaches to epidemiological modeling that consider the impacts of vaccination and provide decision-making support for public health interventions.

AI's Role in Alzheimer’s Research

The research focuses on developing causal models rooted in individual biomarkers to predict patient deterioration in Alzheimer's disease. This involves analyzing various parameters to create personalized treatment models that enhance diagnostic capabilities and improve patient outcomes.

Conclusion

The professor reiterates the importance of collaborative efforts at Purdue University to establish a comprehensive landscape for AI applications specifically within healthcare and engineering domains. It underscores the commitment to leveraging research and education to contribute positively to society, addressing pressing challenges through innovation and interdisciplinary collaboration.

Introduction

Overview of Machine Learning Work

Primary Channel: Provides the main estimation outcomes, which represent the best guess or prediction based on the input data.
Confidence Interval Channel: Offers a measure of uncertainty around the predictions, indicating the reliability of the model's output and the likelihood of various potential outcomes. This dual output system is designed to enhance the robustness and interpretability of machine learning models, allowing for more informed decision-making. The professor emphasizes the importance of incorporating feedback mechanisms based on the confidence interval, which can significantly improve model performance and trust in predictions.

Waves of AI Defined by DARPA

First Wave (1990s): This phase marks the advent of early AI applications characterized by foundational technology, such as chess-playing programs like IBM's Deep Blue, which showcased the potential of AI in strategic thinking and decision-making.
Second Wave (2000s-Present): This era introduced more sophisticated AI systems, epitomized by AlphaGo, which could play Go, a complex board game, at a championship level, defeating world champions and demonstrating advanced learning capabilities.
Third Wave (Present): Focus is shifting towards large language models (LLMs) such as GPT, which have enhanced abilities in natural language processing, learning, and reasoning. Despite these advancements, reasoning skills remain an area that requires further development and refinement.
Fourth Wave (2030 and beyond): The goal of this wave is to achieve artificial general intelligence (AGI), the capacity for AI to perform any cognitive task that a human can undertake, marking a significant milestone in AI evolution.

Applications of AI

AI applications are diverse and impactful, extending to various innovative tasks, including:

Coloring Old Photographs: Utilizing advanced algorithms to restore and enhance historical photos with realistic colors.
Autonomous Driving: Technologies developed for vehicles, with Tesla being a prominent example, showcasing the potential of AI in transportation.
Projects for Social Good: Notable initiatives discussed include:
- Health digital twins designed to target pediatric cardiovascular diseases, creating predictive models for patient care.
- Energy management systems in smart grids that facilitate the integration of renewable energy sources like wind and solar, optimizing energy distribution and consumption.

Research on AI Robustness

AI in Biomedical Applications

A significant project involves the use of AI in conjunction with ultrasound technology to automate heart boundary detection. This project aims to achieve:

Real-Time, High-Resolution Imaging: Enhancing the speed and accuracy of medical imaging processes.
Collaboration with Industry Leaders: Working with companies such as Philips and Fujifilm to bring these advanced imaging techniques to market, affording healthcare professionals better tools for patient diagnosis.

Data Science Education Ecosystem at Purdue

Purdue University fosters a supportive environment for data science education, with initiatives aimed at building data mining communities among undergraduates. Key elements include:

Learning contexts that nurture hands-on collaboration among students and facilitate interaction with industry mentors.
Structured living and learning dormitories designed to cultivate an immersive educational experience focused on data science and machine learning.

Historical Perspectives on AI and ML

AI Development and Models

An overview of key milestones in the evolution of artificial intelligence includes:

1955: Emergence of early AI programs that began exploring reasoning.
1990s: The development of Deep Blue and other pioneering AI technologies.
2012 and Beyond: Breakthrough innovations such as AlexNet, ResNet, AlphaFold, and continuous advancements that push the boundaries of what AI can achieve.

Issues with Deep Learning

Future Directions in AI

Integral AI

Integral AI seeks to extract meaningful scientific knowledge from noisy data, with particular relevance to pandemic-related models such as those centered around COVID-19. This includes:

Unique approaches to epidemiological modeling that consider the impacts of vaccination and provide decision-making support for public health interventions.