This tutorial seeks to showcase AI strategies that provide medical context to patient data with the help of a knowledge graph. This supports personalization through a personalized knowledge graph that captures the patient’s personalized health management objectives within the context of the clinical guidelines and care plan. The continuous capture of this information through the analysis of patient-VHA interactions, and the strategy of creating engaging interactions (conversations) can further augment the personalized knowledge graph. These operations are required to support self-appraisal and self-management, and when necessary perform fail-safe tasks such as connecting the patient to a crisis help-line or professional help. The core innovation is the use of a novel knowledge-infused reinforcement learning method. The by-product of this approach leads to transparency in decision-making with the ability to offer a user understandable explanation.
https://www.knowledgegraph.tech/kgc-2022-tutorial-knowledge-infused-reinforcement-learning/
More Information: https://aiisc.ai/kirl/
Community detection from research papers (AAN dataset) using the algorithms:
K-Means
Louvain
Newman-Girvan
github link to code: https://goo.gl/CXej44
github link to project web page: http://goo.gl/7OOkhI
youtube link to video:https://goo.gl/SCpamf
dropbox link to ppt report video: https://goo.gl/cgACzU
Many powerful Machine Learning algorithms are based on graphs, e.g., Page Rank (Pregel), Recommendation Engines (collaborative filtering), text summarization, and other NLP tasks. Also, the recent developments with Graph Neural Networks connect the worlds of Graphs and Machine Learning even further.
Considering data pre-processing and feature engineering which are both vital tasks in Machine Learning Pipelines extends this relationship across the entire ecosystem. In this session, we will investigate the entire range of Graphs and Machine Learning with many practical exercises.
Community detection from research papers (AAN dataset) using the algorithms:
K-Means
Louvain
Newman-Girvan
github link to code: https://goo.gl/CXej44
github link to project web page: http://goo.gl/7OOkhI
youtube link to video:https://goo.gl/SCpamf
dropbox link to ppt report video: https://goo.gl/cgACzU
Many powerful Machine Learning algorithms are based on graphs, e.g., Page Rank (Pregel), Recommendation Engines (collaborative filtering), text summarization, and other NLP tasks. Also, the recent developments with Graph Neural Networks connect the worlds of Graphs and Machine Learning even further.
Considering data pre-processing and feature engineering which are both vital tasks in Machine Learning Pipelines extends this relationship across the entire ecosystem. In this session, we will investigate the entire range of Graphs and Machine Learning with many practical exercises.
Developing Apps with GPT-4 and ChatGPT_ Build Intelligent Chatbots, Content G...BIHI Oussama
Written in clear and concise language, Developing Apps with GPT-4 and ChatGPT includes easy-to-follow examples to help you understand and apply the concepts to your projects. Python code examples are available in a GitHub repository, and the book includes a glossary of key terms. Ready to harness the power of large language models in your applications? This book is a must.
You'll learn:
The fundamentals and benefits of ChatGPT and GPT-4 and how they work
How to integrate these models into Python-based applications for NLP tasks
How to develop applications using GPT-4 or ChatGPT APIs in Python for text generation, question answering, and content summarization, among other tasks
Advanced GPT topics including prompt engineering, fine-tuning models for specific tasks,
Exploring the Deep Dream Generator (an Art-Making Generative AI) Shalin Hai-Jew
The Deep Dream Generator was created by Google engineer Alexander Mordvintsev in 2014. It has a public facing instance at https://deepdreamgenerator.com/, which enables people to use text prompts and image prompts (individually or in combination) to inspire the art-generating generative AI to output images. This work highlights some process-based walk-throughs of the tool, some practical uses, some lightweight art learning, some aspects of the online social community on this platform, and other insights. Some works by the AI prompted by the presenter may be seen here: https://deepdreamgenerator.com/u/sjjalinn.
(This is the first draft of a slideshow that will be used in a conference later in the year.)
Using Large Language Models in 10 Lines of CodeGautier Marti
Modern NLP models can be daunting: No more bag-of-words but complex neural network architectures, with billions of parameters. Engineers, financial analysts, entrepreneurs, and mere tinkerers, fear not! You can get started with as little as 10 lines of code.
Presentation prepared for the Abu Dhabi Machine Learning Meetup Season 3 Episode 3 hosted at ADGM in Abu Dhabi.
Reinforcement learning:policy gradient (part 1)Bean Yen
The policy gradient theorem is from "Reinforcement Learning : An Introduction". DPG and DDPG is from the original paper.
original link https://docs.google.com/presentation/d/1I3QqfY6h2Pb0a-KEIbKy6v5NuZtnTMLN16Fl-IuNtUo/edit?usp=sharing
Introduction to machine learning. Basics of machine learning. Overview of machine learning. Linear regression. logistic regression. cost function. Gradient descent. sensitivity, specificity. model selection.
Quick introduction to community detection.
Structural properties of real world networks, definition of "communities", fundamental techniques and evaluation measures.
‘Big models’: the success and pitfalls of Transformer models in natural langu...Leiden University
Abstract: Large Language Models receive a lot of attention in the media these days. We have all experienced that generative language models of the GPT family are very fluent and can convincingly answer complex questions. But they also have their limitations and pitfalls. In this presentation I will introduce Transformer-based language models, explain the relation between BERT, GPT, and the 130 thousand other models available on https://huggingface.co. I will discuss their use and applications and why they are so powerful. Then I will point out challenges and pitfalls of Large Language Models and the consequences for our daily work and education.
해당 자료는 풀잎스쿨 18기 중 "설명가능한 인공지능 기획!" 진행 중 Counterfactual Explanation 세션에 대해서 정리한 자료입니다.
논문, Youtube 및 하기 자료를 바탕으로 정리되었습니다.
https://christophm.github.io/interpretable-ml-book/
Artificial Intelligence is increasingly playing an integral role in determining our day-to-day experiences. Moreover, with proliferation of AI based solutions in areas such as hiring, lending, criminal justice, healthcare, and education, the resulting personal and professional implications of AI are far-reaching. The dominant role played by AI models in these domains has led to a growing concern regarding potential bias in these models, and a demand for model transparency and interpretability. In addition, model explainability is a prerequisite for building trust and adoption of AI systems in high stakes domains requiring reliability and safety such as healthcare and automated transportation, and critical industrial applications with significant economic implications such as predictive maintenance, exploration of natural resources, and climate change modeling.
As a consequence, AI researchers and practitioners have focused their attention on explainable AI to help them better trust and understand models at scale. The challenges for the research community include (i) defining model explainability, (ii) formulating explainability tasks for understanding model behavior and developing solutions for these tasks, and finally (iii) designing measures for evaluating the performance of models in explainability tasks.
In this tutorial, we present an overview of model interpretability and explainability in AI, key regulations / laws, and techniques / tools for providing explainability as part of AI/ML systems. Then, we focus on the application of explainability techniques in industry, wherein we present practical challenges / guidelines for effectively using explainability techniques and lessons learned from deploying explainable models for several web-scale machine learning and data mining applications. We present case studies across different companies, spanning application domains such as search & recommendation systems, sales, lending, and fraud detection. Finally, based on our experiences in industry, we identify open problems and research directions for the data mining / machine learning community.
Unsupervised learning representation with Deep Convolutional Generative Adversarial Network, Paper by Alec Radford, Luke Metz, and Soumith Chintala
(indico Research, Facebook AI Research).
And then there were ... Large Language ModelsLeon Dohmen
It is not often even in the ICT world that one witnesses a revolution. The rise of the Personal Computer, the rise of mobile telephony and, of course, the rise of the Internet are some of those revolutions. So what is ChatGPT really? Is ChatGPT also such a revolution? And like any revolution, does ChatGPT have its winners and losers? And who are they? How do we ensure that ChatGPT contributes to a positive impulse for "Smart Humanity?".
During a key note om April 3 and 13 2023 Piek Vossen explained the impact of Large Language Models like ChatGPT.
Prof. PhD. Piek Th.J.M. Vossen, is Full professor of Computational Lexicology at the Faculty of Humanities, Department of Language, Literature and Communication (LCC) at VU Amsterdam:
What is ChatGPT? What technology and thought processes underlie it? What are its consequences? What choices are being made? In the presentation, Piek will elaborate on the basic principles behind Large Language Models and how they are used as a basis for Deep Learning in which they are fine-tuned for specific tasks. He will also discuss a specific variant GPT that underlies ChatGPT. It covers what ChatGPT can and cannot do, what it is good for and what the risks are.
Data/AI driven product development: from video streaming to telehealthXavier Amatriain
Healthcare is different from any other application domain, or is it not? While it is true that there are specific aspects, such as high stakes decisions and a complex regulatory framework, that make healthcare somewhat different, it is also the case that many of the lessons learned from building data-driven products in other domains translate remarcably well into healthcare. This is particularly so because healthcare is also a user facing domain, where users can be both patients or healthcare professionals. Given that data has shown to improve user experience while ensuring quality and scalability, few would argue that healthcare cannot benefit from being much more data-driven than it has traditionally been.
In this talk, I described how this experience building impactful data and AI solutions into user facing products for decades can be leveraged to revolutionize telehealth. At Curai, we combine approaches such as state-of-the-art large language models with expert systems in areas such as NLP, vision, and automated diagnosis to augment and scale doctors, and to improve user experience and healthcare outcomes. We will see some of those applications while analyzing the role of data and ML algorithms in making them possible.
Developing Apps with GPT-4 and ChatGPT_ Build Intelligent Chatbots, Content G...BIHI Oussama
Written in clear and concise language, Developing Apps with GPT-4 and ChatGPT includes easy-to-follow examples to help you understand and apply the concepts to your projects. Python code examples are available in a GitHub repository, and the book includes a glossary of key terms. Ready to harness the power of large language models in your applications? This book is a must.
You'll learn:
The fundamentals and benefits of ChatGPT and GPT-4 and how they work
How to integrate these models into Python-based applications for NLP tasks
How to develop applications using GPT-4 or ChatGPT APIs in Python for text generation, question answering, and content summarization, among other tasks
Advanced GPT topics including prompt engineering, fine-tuning models for specific tasks,
Exploring the Deep Dream Generator (an Art-Making Generative AI) Shalin Hai-Jew
The Deep Dream Generator was created by Google engineer Alexander Mordvintsev in 2014. It has a public facing instance at https://deepdreamgenerator.com/, which enables people to use text prompts and image prompts (individually or in combination) to inspire the art-generating generative AI to output images. This work highlights some process-based walk-throughs of the tool, some practical uses, some lightweight art learning, some aspects of the online social community on this platform, and other insights. Some works by the AI prompted by the presenter may be seen here: https://deepdreamgenerator.com/u/sjjalinn.
(This is the first draft of a slideshow that will be used in a conference later in the year.)
Using Large Language Models in 10 Lines of CodeGautier Marti
Modern NLP models can be daunting: No more bag-of-words but complex neural network architectures, with billions of parameters. Engineers, financial analysts, entrepreneurs, and mere tinkerers, fear not! You can get started with as little as 10 lines of code.
Presentation prepared for the Abu Dhabi Machine Learning Meetup Season 3 Episode 3 hosted at ADGM in Abu Dhabi.
Reinforcement learning:policy gradient (part 1)Bean Yen
The policy gradient theorem is from "Reinforcement Learning : An Introduction". DPG and DDPG is from the original paper.
original link https://docs.google.com/presentation/d/1I3QqfY6h2Pb0a-KEIbKy6v5NuZtnTMLN16Fl-IuNtUo/edit?usp=sharing
Introduction to machine learning. Basics of machine learning. Overview of machine learning. Linear regression. logistic regression. cost function. Gradient descent. sensitivity, specificity. model selection.
Quick introduction to community detection.
Structural properties of real world networks, definition of "communities", fundamental techniques and evaluation measures.
‘Big models’: the success and pitfalls of Transformer models in natural langu...Leiden University
Abstract: Large Language Models receive a lot of attention in the media these days. We have all experienced that generative language models of the GPT family are very fluent and can convincingly answer complex questions. But they also have their limitations and pitfalls. In this presentation I will introduce Transformer-based language models, explain the relation between BERT, GPT, and the 130 thousand other models available on https://huggingface.co. I will discuss their use and applications and why they are so powerful. Then I will point out challenges and pitfalls of Large Language Models and the consequences for our daily work and education.
해당 자료는 풀잎스쿨 18기 중 "설명가능한 인공지능 기획!" 진행 중 Counterfactual Explanation 세션에 대해서 정리한 자료입니다.
논문, Youtube 및 하기 자료를 바탕으로 정리되었습니다.
https://christophm.github.io/interpretable-ml-book/
Artificial Intelligence is increasingly playing an integral role in determining our day-to-day experiences. Moreover, with proliferation of AI based solutions in areas such as hiring, lending, criminal justice, healthcare, and education, the resulting personal and professional implications of AI are far-reaching. The dominant role played by AI models in these domains has led to a growing concern regarding potential bias in these models, and a demand for model transparency and interpretability. In addition, model explainability is a prerequisite for building trust and adoption of AI systems in high stakes domains requiring reliability and safety such as healthcare and automated transportation, and critical industrial applications with significant economic implications such as predictive maintenance, exploration of natural resources, and climate change modeling.
As a consequence, AI researchers and practitioners have focused their attention on explainable AI to help them better trust and understand models at scale. The challenges for the research community include (i) defining model explainability, (ii) formulating explainability tasks for understanding model behavior and developing solutions for these tasks, and finally (iii) designing measures for evaluating the performance of models in explainability tasks.
In this tutorial, we present an overview of model interpretability and explainability in AI, key regulations / laws, and techniques / tools for providing explainability as part of AI/ML systems. Then, we focus on the application of explainability techniques in industry, wherein we present practical challenges / guidelines for effectively using explainability techniques and lessons learned from deploying explainable models for several web-scale machine learning and data mining applications. We present case studies across different companies, spanning application domains such as search & recommendation systems, sales, lending, and fraud detection. Finally, based on our experiences in industry, we identify open problems and research directions for the data mining / machine learning community.
Unsupervised learning representation with Deep Convolutional Generative Adversarial Network, Paper by Alec Radford, Luke Metz, and Soumith Chintala
(indico Research, Facebook AI Research).
And then there were ... Large Language ModelsLeon Dohmen
It is not often even in the ICT world that one witnesses a revolution. The rise of the Personal Computer, the rise of mobile telephony and, of course, the rise of the Internet are some of those revolutions. So what is ChatGPT really? Is ChatGPT also such a revolution? And like any revolution, does ChatGPT have its winners and losers? And who are they? How do we ensure that ChatGPT contributes to a positive impulse for "Smart Humanity?".
During a key note om April 3 and 13 2023 Piek Vossen explained the impact of Large Language Models like ChatGPT.
Prof. PhD. Piek Th.J.M. Vossen, is Full professor of Computational Lexicology at the Faculty of Humanities, Department of Language, Literature and Communication (LCC) at VU Amsterdam:
What is ChatGPT? What technology and thought processes underlie it? What are its consequences? What choices are being made? In the presentation, Piek will elaborate on the basic principles behind Large Language Models and how they are used as a basis for Deep Learning in which they are fine-tuned for specific tasks. He will also discuss a specific variant GPT that underlies ChatGPT. It covers what ChatGPT can and cannot do, what it is good for and what the risks are.
Data/AI driven product development: from video streaming to telehealthXavier Amatriain
Healthcare is different from any other application domain, or is it not? While it is true that there are specific aspects, such as high stakes decisions and a complex regulatory framework, that make healthcare somewhat different, it is also the case that many of the lessons learned from building data-driven products in other domains translate remarcably well into healthcare. This is particularly so because healthcare is also a user facing domain, where users can be both patients or healthcare professionals. Given that data has shown to improve user experience while ensuring quality and scalability, few would argue that healthcare cannot benefit from being much more data-driven than it has traditionally been.
In this talk, I described how this experience building impactful data and AI solutions into user facing products for decades can be leveraged to revolutionize telehealth. At Curai, we combine approaches such as state-of-the-art large language models with expert systems in areas such as NLP, vision, and automated diagnosis to augment and scale doctors, and to improve user experience and healthcare outcomes. We will see some of those applications while analyzing the role of data and ML algorithms in making them possible.
3282016 Additional Book Resourceshttpscourserooma.cap.docxtamicawaysmith
3/28/2016 Additional Book Resources
https://courserooma.capella.edu/bbcswebdav/institution/ITFP/ITFP3300/Version0715/Course_Files/cf_additional_book_resources.html 1/2
To conduct additional research, you may search your local library or bookstore for the following course
related books:
BagtesBrkljac, N. (2012). Computer science, technology and applications: Virtual reality. Hauppage, NY:
Nova Science Publishers.
Crandall, B., Klein, G., & Hoffman, R. R. (2006). Working minds: A practitioner's guide to cognitive task
analysis. Cambridge, MA: MIT Press.
Dautenhahn, K., Bond, A. H., & Cañamero, L. (2002). Socially intelligent agents: Creating relationships
with computers and robots. Hingham, MA: Kluwer Academic Publishers.
Emerald Publishing Group. (2005). Digital library usability studies. Bradford, UK: Emerald Group
Publishing.
Fowler, S., & Stanwick, V. (2004). Interactive technologies: Web application design handbook: Best
practices for webbased software. Burlington, MA: Morgan Kaufmann.
Hillis, K. (1999). Digital sensations: Space, identity, and embodiment in virtual reality. Minneapolis, MN:
University of Minnesota Press.
Hashimoto, A. (2003). Visual design fundamentals: A digital approach. Irvine, CA: Delmar Cengage
Learning.
Holland, J. M. (2003). Designing autonomous mobile robots: Inside the mind of an intelligent machine.
Burlington, MA: Newnes Publishing.
Leung, L. (2008). Digital experience design: Ideas, industries, interaction. Bristol, UK: Intellect Ltd
Publishers.
Mavor, A. S., & Durlach, N. I. (Eds.). (1994). Virtual reality: Scientific and technological challenges.
Washington, DC: National Academies Press.
Proctor, R. W., & KimPhuong, L. V. (2004). Handbook of human factors in web design. Boca Raton, FL:
CRC Press.
Salvendy, G. (2012). Handbook of human factors and ergonomics. (4th ed.). Hoboken, NJ: John Wiley &
Sons.
Sherman, P. (2006). Usability success stories: How organizations improve by making easiertouse software
and web sites. Burlington, VT: Ashgate Publishing Company.
Steinfeld, E., & Maisel, J. L. (2012). Universal design: Creating inclusive environments. Hoboken, NJ:
John Wiley & Sons.
Westwood, J. D., Haluck, R. S., & Hoffman, H. M. (2007). Studies in health technology and informatics:
Medicine meets virtual reality. Amsterdam, Netherlands: IOS Press.
Print
Additional Book Resources
javascript:window.print()
3/28/2016 Additional Book Resources
https://courserooma.capella.edu/bbcswebdav/institution/ITFP/ITFP3300/Version0715/Course_Files/cf_additional_book_resources.html 2/2
Woolgar, S. (2002). Virtual society?: Technology, cyberbole, reality. Oxford, UK: Oxford University Press.
Designing a Complete Network Security Policy
Learning Outcomes: At the end of the assignment, student should be able:
· To have an understanding of the network security issues in organizations and how to solve them by developing and applying a network security policy, which contains different security ...
AI TESTING: ENSURING A GOOD DATA SPLIT BETWEEN DATA SETS (TRAINING AND TEST) ...ijsc
Artificial Intelligence and Machine Learning have been around for a long time. In recent years, there has been a surge in popularity for applications integrating AI and ML technology. As with traditional development, software testing is a critical component of a successful AI/ML application. The development methodology used in AI/ML contrasts significantly from traditional development. In light of these distinctions, various software testing challenges arise. The emphasis of this paper is on the challenge of effectively splitting the data into training and testing data sets. By applying a k-Means clustering strategy to the data set followed by a decision tree, we can significantly increase the likelihood of the training data set to represent the domain of the full dataset and thus avoid training a model that is likely to fail because it has only learned a subset of the full data domain.
The Skynet Effect: How HR Can Best Utilize AIAggregage
https://www.humanresourcestoday.com/frs/24235077/the-skynet-effect--how-hr-can-best-utilize-ai/email
AI this, AI that. No matter where you go, AI seems to be all anyone in HR wants to talk about. It might be a little irritating, but it’s inescapable for a good reason. Artificial intelligence, specifically ChatGPT, is now an important topic of conversation for all industries.
Like anything new, there are plenty of questions and misconceptions about how AI will change workplace dynamics. But modern AI isn’t Skynet trying to take over the world, and instead of fearing it, your organization can embrace the efficiencies and positive impacts that it offers.
In this webinar, Iveta Brigis, Vice President, People Operations, and Wesley Pasfield, Head of Data Science will address:
• How to better understand AI, specifically ChatGPT
• Use cases for AI in the employer space
• Considerations and questions to ask when evaluating an AI vendor
• Concerns about AI affecting HR employment availability
• How AI can enhance HR jobs when used responsibly
Essay about kindnessChoose any three consecutive days. During th.docxSALU18
Essay about kindness
Choose any three consecutive days. During that period, practice absolute kindness toward any and all with whom you interact. Go out of your way to be courteous and thoughtful. Identify specific individuals, both close to you and interpersonally more distant, and identify acts of kindness that you can do for them—and then perform those various acts of kindness. During this period of time you are to say and do no unkind thing. Rather, act for these three days as if you were a Gandhi, a Jesus, a Socrates, a St. Francis of Assisi, or a Mother Theresa. Then, write a short 3- page essay describing the lessons you learned, how they may relate to anything you have read thus far, and what commitments—if any—you have consequently made because of what you have learned.
PS: Let’s say I am a 20-year old female
ISE 510 Jones & Bartlett (JBL) Lecture Presentation and Assignment Guidelines and Rubric
Prompt: There are four activities in this course that require you to log in to the Jones & Bartlett Learning website, where you will review a lecture presentation
and complete an assignment. These activities occur in Modules One, Two, Six, and Eight. For each activity, you will download a Microsoft Word document from
JBL, fill out your answers directly in the Word document, then submit the completed document to Blackboard. Navigate to your Jones & Bartlett Learning Lecture
Presentation and Assignment website here.
Review the lecture presentation and complete the assignments listed in the following modules:
Module One (Task 1-3):
o Review Lecture Presentation: Risk Management Fundamentals
o Complete Assignment: Application of Risk Management Techniques
Module Two (Task 2-2):
o Review Lecture Presentation: Compliance Laws, Standards, and Best Practices
o Complete Assignment: PCI DSS and the Seven Domains
Module Six (Task 6-2):
o Review Lecture Presentation: Structuring a Computer Incident Response Team and Plan
o Complete Lab Manual: Create a CIRT Response Plan for a Typical IT Infrastructure
Module Eight (Task 8-2):
o Review the Lecture Presentation: Strategies for Mitigating Risk
o Complete Lab Manual: Developing a Risk-Mitigation Plan Outline for an IT Infrastructure
Note: When you purchase your course material bundle, you will receive an email from Jones & Bartlett Learning with your access to these materials. Information
on purchasing your course material bundle is located in the course syllabus.
Specifically, the following critical elements must be addressed:
Accuracy of responses
Integration and application of concepts
Articulation of responses
Guidelines for Submission: Follow the submission guidelines laid out in the Jones & Bartlett Learning assignment prompt. For each activity, you will download a
Microsoft Word document from JBL, fill out your answers directly in the Word document, then submit the completed document to Blackboard. Any sources must
be cited a ...
Paper OneLength- 1000- 1200 words- 3-5-4 pages- exclusive of the Work.docxestefana2345678
Paper One
Length : 1000- 1200 words/ 3.5-4 pages, exclusive of the Work Cited page
For your first paper, you’ll be analyzing impediments to your own critical thinking and how they shaped your decision making in a specific decision in your life.
First, think back to a decision you’ve made that you either now see as a bad decision or that you’re still not fully sure you thought through critically. It doesn’t have to be a super-personal decision (why did I date that girl for so long in high school?), and it can even be a decision that’s had a good outcome (why did I choose this university?), as long as you can express how you didn’t really think critically about it at the time .
Elements of Reasoning : For prewriting purposes, go around the circle of elements with this decision as you made it then , paying close attention to who you were when you made it (your Point of View). You’ll want to use these notes as you describe your decision-making process and put any elements of reasoning in bold in your paper. Plan to use between 3-5 elements of reasoning in your paper.
Impediments : Finally, think about what types of impediments got in your way as you made this decision. Develop paragraphs in your paper around these impediments and put them in bold type in your paper as well. Your thesis statement should say something about how the impediments that blocked your critical thinking interacted with the elements of reasoning to keep you from using them effectively.
Because all papers in CRTW must include documented material, make sure you quote Nosich at least once when talking about at least one impediment that hindered your critical thinking. You'll then need to give the page number in MLA format in your paper and create a Works Cited page with Nosich's book on it at the end of your paper.
Hints:
1. Don’t be afraid to use the first person “I.†This is a paper about you and your thinking.
2. You can tell this as a story if you’d like, so long as it’s clear that you’re analyzing your own thinking and which impediments and elements of reasoning were (or weren’t) involved.
3. Whether you tell this as a story or write it as a more formal academic paper, your introduction should give some context to your decision: when was it, what was it, and why did you need to make it?
4. In your conclusion, rather than repeating what you’ve already said in the introduction and body of the paper, please try to reflect on what you’ve learned from analyzing this decision and/or making the decision in the first place. What might you do differently in the future? How might you approach the same impediment(s) if you feel them creeping into your thought process in future decisions?
Proprietary
KATHY FORSYTH
CAPELLA UNIVERSITY
Tele Psych Staff Training Session
Proprietary
What is Telehealth
and Tele psych?
ïµ Technology ïµ Four models
ïµ Closing the gap access to healthcare ïµ Care when they need it, no matter the
distance
ïµ Decreasing cost.
Week 5 Reflection Pulse checkTop of FormBottom of FormPulse .docxhelzerpatrina
Week 5 Reflection Pulse check
Top of Form
Bottom of Form
Pulse Check
The “Pulse” Check. Where are you in your journey and how are you doing?
DQ2 of Week 5 is an opportunity for you to self-assess and reflect on your journey. You can write about it expressing your navigation through the Role course. Model from the scripts in this week but be specific to your experience.
Evaluate how you have achieved course competencies and your plans to develop further in these areas. The course competencies for this course are as follows:
1. Explore the historical evolution of the advance practice nurse.
2. Differentiate the roles and scope of practice for nurses working in advanced clinical, education, administration, informatics, research, and health policy arenas.
3. Analyze attributes of the practice arena such as access and availability, degree of consumer choice, competition, and financing that impact advanced practice nurses and their ability to effectively collaborate with other health professionals.
4. Integrate evidence from research and theory into discussions of practice competencies, health promotion and disease prevention strategies, quality improvement, and safety standards.
5. Identify collaborative, organizational, communication, and leadership skills in working with other professionals in healthcare facilities and/or academic institutions.
6. Synthesize knowledge from values theory, ethics, and legal/regulatory statutes in the development of a personal philosophy for a career as an advanced practice nurse.
Guidelines: Support your responses with scholarly academic references using APA style format. Assigned course readings and online library resources are preferred. Weekly lecture notes are designed as overviews to the topic for the respective week and should not serve as a citation or reference.
In your discussion question response, provide a substantive response that illustrates a well-reasoned and thoughtful response; is factually correct with relevant scholarly citations, references, and examples that demonstrate a clear connection to the readings.
Week 5
Article and Videos
Read the following peer reviewed article. Incorporate key points in your DQ 1 and any other assignment.
A Day in the Life: Nurse Johnson & Johnson Nursing Campaign Mystery Box Unboxing ( https://www.youtube.com/watch?v=rgojAOyPvfk&list=PLU05He9EuwhlRCO5iVPgjeA7-K1lE3Y8X )
Strech, S., & Wyatt, D. A. (2013). Partnering to lead change: Nurses' role in the redesign of health care. Association of Operating Room Nurses.AORN Journal, 98(3), 260-6. doi:http://dx.doi.org.southuniversity.libproxy.edmc.edu/10.1016/j.aorn.2013.07.006
A Day in the Life: Nurse Educator A Day in the Life - Susan, Nurse Educator, MSN, RN, Ph.D. Candidate ( https://www.youtube.com/watch?v=Z_sG4GRtP-o )
A Day in the Life: Family Nurse Practitioner A Day in the Life - Steve (Family Nurse Practitioner) (https://www.youtube.com/watch?v=d-kL1OFbCC8 )
A Day in the Life: Palliative Care NP A Day in ...
Depression Detection in Tweets using Logistic Regression Modelijtsrd
In the growing world of modernization, mental health issues like depression, anxiety and stress are very normal among people and social media like Facebook, Instagram and Twitter have boosted the growth of such mental health. Everything has its legitimacy and negative mark. During this pandemic, people are more likely to suffer from mental health issues, they are available 24 7 and are cut off from the real world. Past examinations have shown that individuals who invest more energy via online media are bound to be depressed. In this project, we find out people who are depressed based on their tweets, followers, following and many other factors. For this, I have trained and tested our text classifier, which will distinguish between the user who is depressed or not depressed. Rahul Kumar Sharma | Vijayakumar A "Depression Detection in Tweets using Logistic Regression Model" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-5 | Issue-4 , June 2021, URL: https://www.ijtsrd.compapers/ijtsrd41284.pdf Paper URL: https://www.ijtsrd.comcomputer-science/data-miining/41284/depression-detection-in-tweets-using-logistic-regression-model/rahul-kumar-sharma
This presentation, created by Syed Faiz ul Hassan, explores the profound influence of media on public perception and behavior. It delves into the evolution of media from oral traditions to modern digital and social media platforms. Key topics include the role of media in information propagation, socialization, crisis awareness, globalization, and education. The presentation also examines media influence through agenda setting, propaganda, and manipulative techniques used by advertisers and marketers. Furthermore, it highlights the impact of surveillance enabled by media technologies on personal behavior and preferences. Through this comprehensive overview, the presentation aims to shed light on how media shapes collective consciousness and public opinion.
Collapsing Narratives: Exploring Non-Linearity • a micro report by Rosie WellsRosie Wells
Insight: In a landscape where traditional narrative structures are giving way to fragmented and non-linear forms of storytelling, there lies immense potential for creativity and exploration.
'Collapsing Narratives: Exploring Non-Linearity' is a micro report from Rosie Wells.
Rosie Wells is an Arts & Cultural Strategist uniquely positioned at the intersection of grassroots and mainstream storytelling.
Their work is focused on developing meaningful and lasting connections that can drive social change.
Please download this presentation to enjoy the hyperlinks!
Mastering the Concepts Tested in the Databricks Certified Data Engineer Assoc...SkillCertProExams
• For a full set of 760+ questions. Go to
https://skillcertpro.com/product/databricks-certified-data-engineer-associate-exam-questions/
• SkillCertPro offers detailed explanations to each question which helps to understand the concepts better.
• It is recommended to score above 85% in SkillCertPro exams before attempting a real exam.
• SkillCertPro updates exam questions every 2 weeks.
• You will get life time access and life time free updates
• SkillCertPro assures 100% pass guarantee in first attempt.
1. Knowledge Infused
Reinforcement Learning
Use-case: Conversational Systems
Spread the word!
Tweet, post on LinkedIn your insights, screenshots, and questions
#KGC2022 #ProcessKnowledge #KnowledgeInfusedLearning #safeAI
#InterpretableML #ExplainbleAI
http://aiisc.ai/kirl
2. About the Tutors
2
Qi Zhang,
Professor, CS@UofSC
Planning and Reinforcement
Learning in Multi-Agent Systems,
Knowledge-infused
Reinforcement Learning
Amit Sheth,
Professor, CS@UofSC
Director, AI Institute @ UofSC
Knowledge-infused Learning,
Knowledge Graphs, Semantic
Web, Natural Language
Understanding,
Conversational Systems
Manas Gaur,
Ph.D. in CS @ UofSC
Researcher, AI Institute @ UofSC
Knowledge-infused Learning,
Natural Language
Understanding, Knowledge
Graphs,
Kaushik Roy,
Ph.D. in CS @ UofSC
Researcher, AI Institute @ UofSC
Knowledge-infused
Reinforcement Learning,
Conversational Systems
3. D
O
M
A
I
N
S
3
Conversational AI in [Healthcare]
Sheth et al. 2018, Roy et al. 2021, Gaur et al. 2022
[Pandemic/Crisis] Matching Support Seekers with
Support Providers (Gaur et al. 2021)
[Education] ALLURE Collaborative Rubik’s Cube Solver
Agostinelli et al. 2021 (https://tinyurl.com/AIISC-Rubiks)
[Food] Allergy-Aware Recipe Understanding
(Khandelwal et al. 2022) (https://tinyurl.com/AIISC-Allergies)
4. How to Build a Conversational Agent
4
Attention Neural Network
(Vaswani et al. NIPS’17)
Large Amount of Textual Conversational
Data in General Domain
Trained Generative Model
Picture Credit: https://keyreply.com/conversational-ai-
for-healthcare/
5. Agent tasked to Converse with Patient with Mental
healthcare Condition
5
Do you feel nervous?
More than half the days
Do you feel irritated or self
destructive?
Do you feel something
extreme might happen to
you?
Are you able to relax?
The model can
generate and ask
either of these
questions
They are either bad
questions or irrelevant
( A clinician won’t ask
either of these)
A model trained to asked questions
Risky
6. 6
Unsafe Question Generation
Do you feel nervous?
More than half the days
Do you feel irritated or self
destructive?
Do you feel something
extreme might happen to
you?
Are you able to relax?
The model can
generate and ask
either of these
questions
A model trained to asked questions
7. Large Language Models Hallucinate
7
❏ Generate Factually Incorrect Response
❏ Blenderbot 1 [Roller et al. 2020]
❏ Blenderbot 2 [https://tinyurl.com/Meta-bbot2]
❏ Generate Questions that aren’t Safe
❏ OpenAI GPT-3 [Brown et al. 2020]
❏ DeepMind’s RETRO [Borgeaud et al. 2021]
❏ Google’s LaMDA [Thoppilan et al. 2022]
❏ Utilize Google’s long list of Safety Guidelines to place constraint on
Safety.
8. 8
Do you feel nervous?
More than half the days
Do you feel irritated or self
destructive?
Do you feel something
extreme might happen to
you?
Are you able to relax?
Do you feel nervous?
More than half the days
Do you feel Irritated?
Are you bothered by
becoming easily annoyed
or irritable?
Are you bothered by any
relaxation troubles?
Knowledge
Infusion using
Medical
Questionnaire
(MedQ)
These questions are
medically valid and safe.
Safety
Checks
Knowledge fix : Safety Checks
Roy and Gaur et al. ACL’22
9. 9
Do you feel nervous?
More than half the days
T5
(Raffel et.
al.
ACL’20)
KI
Attention
Model
(Ours)
30.6%
17.1%
10.6%
13.3%
KI: Knowledge Infusion
Do you feel Irritated?
Are you bothered by
becoming easily annoyed
or irritable?
Are you bothered by any
relaxation troubles?
(Ours) Roy and Gaur et al. ACL
(under review)
Effect of Knowledge fix : Safety Checks
10. Attention Model
I feel bothered by little interest
and have least pleasure in
doing anything
Did you check your dopamine
levels?
Do you feel your brain is
affected?
Did you intend to indulge in
risky behaviors
I feel bothered by little interest
and have least pleasure in
doing anything
What does “lack of pleasure”
mean to you?
Do you feel little pleasure doing
things you used to enjoy?
How long have you struggled with
lack of interest in things you used
to enjoy?
KI Attention Model
DSM-5
Lexicons for
Depression
SCID for
Depression
PHQ-9
Questionnair
e
Gaur et al. WWW’19; Yazdavar et al. ASONAM’17
0.83
0.78
0.71
0.33
0.21
0.19
11. Safety Checks
11
If is cause then symptom
If is symptom then medication
If is medication then treatment
Probability next question generation is
Process
Do you feel nervous?
More than half the days
Do you feel Irritated?
Are you bothered by
becoming easily annoyed
or irritable?
Are you bothered by any
relaxation troubles?
12. Uncertainty and Risk
12
Generation Task
Handling Uncertainty or Risk
Human
Annotation
Experience
(e.g. History)
Web Search
Corpus
Expert
Guidelines
Generative
Output
Classification
Output
Labeled Dataset
13. Safety Checks: A method of infuse Knowledge
13
Generation Task
Handling Uncertainty or Risk
Generative
Output
Classification
Output
Labeled Dataset
Information
graph (KG,
Lexicons,
etc.)
Kashyap and
Sheth CIKM’94
14. Viable forms of knowledge to infuse
14
Generative
Output
Classification
Output
Labeled Dataset
Interpretability
User-level Explainability
Information
graph (KG,
Lexicons,
etc.)
Kashyap and
Sheth CIKM’94
Generative
Output
Classification
Output
Labeled Dataset
Interpretability
User-level Explainability
ConceptNet
Healthcare
lexicons
15. 15
Example KG constructed either from manual effort (A, B, C), automatically (D, E), or semi-automatically (F)
(A) is empathi ontology designed to
identify concepts in disaster
scenarios (Gaur et al. 2019).
(B) Chem2Bio2RDF (Chen et al. 2010).
(C) ATOMIC (Sap et al. 2019).
(D) Education Knowledge Graph by
Embibe (Faldu et al. 2020).
(E) Event Cascade Graph in WildFire
(Jiang et al. 2019).
(F) Opioid Drug Knowledge Graph
(Kamdar et al. 2019)
16. Process Knowledge : Suicide Risk
16
Columbia Suicide Severity
Rating Scale (C-SSRS)
Process Knowledge
Structure of C-SSRS
Posner et al. 2011
Am. Journal of Psychiatry
17. Process Knowledge : Anxiety
17
Spitzer et al. 2006
Archives of Internal Medicine
Bartolo et al. 2017
SciELO Brasil
18. Knowledge Infused Learning (KiL)
18
Knowledge-infused Learning is a class of Neuro-Symbolic AI
techniques that incorporate broader forms of knowledge (lexical,
domain-specific, common-sense, and constraint-based) into
addressing limitations of either symbolic or statistical AI approaches,
such as model interpretations and user-level explanations.
Compared to powerful statistical AI that exploit data, KiL benefit
from data as well as knowledge.
19. 19
Shallow Infusion of Process Knowledge
Deep Language
Model for
Classification
Semantic
Embedding Loss
❏ External Knowledge is utilized in Annotating the
Dataset
❏ Patient Health Questionnaire-9 (PHQ-9): an
instrument to measure severity of Depression.
❏ Input is matched to Question
❏ To perform user-level classification
❏ Encodes knowledge as either word embeddings (text)
or graph embeddings (graphs,rules, trees)
❏ User-level Explainability and Safety:
❏ Model’s classification is checked by the questions
answered correctly.
❏ Correctly answered questions are matched to
concepts highlighted in the input.
Agarwal and Gupta et al. ACL RR
Under Review
embedding
20. Semi-Deep Infusion of Process Knowledge
20
Deep Language Model
For Representation
Bernoulli Loss
embedding
Decision
Tree
❏ Utilizes knowledge in its original form to modify the
parameters of the ML model
❏ We employ and adapt interpretable ML models
E.g. Decision Trees, Naive Bayes, K-Nearest Neighbors
❏ Knowledge is incorporated in the objective function to
achieve parameter tweaking.
❏ User-level Explainability, Safety, & Interpretability:
❏ Model’s outcome is a tree → Interpretability
❏ Predicted Tree is checked with PHQ-9 Questions
→ User Level Explainability
❏ Bernoulli Loss keep refining the tree and can be
tweaked by experts.
Roy and Gaur et al. Arxiv
(https://tinyurl.com/AIISC-PKiL)
21. Deep Infusion of Process Knowledge
21
❏ Every layer within the deep neural network is checked for
information divergence using external knowledge
❏ In case of Process Knowledge and in the context of
Depression, we use Structured Clinical Interview
information for Depression
❏ Knowledge Control:
❏ Checks if current learned representation can answer
questions in SCID
❏ If none questions are answered,
❏ Information from SCID replace learned
representation
❏ If some questions are answered incorrectly ( )
❏ Learned representation is altered as follows:
Structured Clinical
Interview for Depression
(SCID)
backprop
Kursuncu and Gaur et al. AAAI-MAKE’19
22. 22
Context Sensitive Capture:
Statistical AI is opinionated based on the text it sees
and input is partial representation of the world.
Uncertainty and Risk:
Statistical AI, fail to establish the connection between
input and output
User-level Explainable:
Statistical AI’s explanations are system-oriented
and not rich enough for user-level understanding.
Interpretable:
A Statistical AI model that you can understand and
control
Task Transferable:
Statistical AI learns the data and not the task
Knowledge can highlight the context in input.
Knowledge can assess risky prediction
Knowledge can influence attention of statistical AI.
Knowledge can enable User-level explanations
Knowledge can help in generalize across tasks
Benefits of Knowledge-infused Learning
23. Tutorial Overview
23
❏ Challenges in Conversational Agents in Healthcare
❏ Can a model capture context and handle uncertainty in input?
❏ Can user-level explanations be obtained from the success or failure of an AI model?
❏ Can we control an AI model by making it interpretable?
❏ How can we make an AI model self-explainable?
❏ Tutorial Highlights
❏ Reinforcement Learning for Co-operative Multi-Agent System Interactions
❏ Process Knowledge Integration in Reinforcement Learning
❏ Demonstration of Process Knowledge Infusion
28. What’s Reinforcement Learning?
Learn to make good sequences of decisions
⬢ Repeated interactions with environment
⬢ Reward to measure the quality of sequence of decisions
28
29. What’s Reinforcement Learning?
Learn to make good sequences of decisions
⬢ Repeated interactions with environment
⬢ Reward to measure the quality of sequence of decisions
⬢ Don’t know in advance how environment works
29
30. Animal Reinforcements
Negative reinforcements
⬢ Pain and hunger
Positive reinforcements
⬢ Pleasure and food
30
Reinforcements used to train animals
Let’s do the same with machines!
⬢ Here reinforcements are called rewards.
⬢ Machines are called (reinforcement learning) agents.
31. Rewards
A reward r_t is a scalar feedback signal
Indicates how well the agent is doing at step t
The agent’s job is to maximize cumulative reward
31
Reinforcement learning is based on the reward hypothesis:
All goals can be described by the maximization of expected cumulative reward
32. Examples of Rewards
Defeat the world champion at Chess
⬢ +/- reward for winning/losing a game
⬢ No reward (i.e. 0 reward) in the middle of a game
Manage an investment portfolio
⬢ + reward for each $ in bank
Control a power station
⬢ + reward for producing power
⬢ − reward for exceeding safety thresholds
Make a humanoid robot walk
⬢ + reward for forward motion
⬢ − reward for falling over
32
33. The RL Loop
⬢ Goal: select actions to maximize total future reward
⬢ May require sacrificing immediate reward to gain more long-term reward
33
Environment
Agent
Observation,
Reward
Action
Examples: game playing (Go, Atari), operations research (pricing,
vehicle routing), robotic control, conversational agents, autonomous
vehicles, computational finance, etc.
34. The RL Loop: Blood Pressure Control
34
Environment
Agent
Observation:
Patient’s information
Reward:
+1 if healthy pressure; -
0.05 for side effects
Action:
Exercise or medication
⬢ Goal: select actions to maximize total future reward
⬢ May require sacrificing immediate reward to gain more long-term reward
35. The RL Loop: Conversational Agent
35
Environment
Agent
Observation:
User utterance
Reward:
Task completion, user
satisfaction, etc.
Action:
Next utterance
⬢ Goal: select actions to maximize total future reward
⬢ May require sacrificing immediate reward to gain more long-term reward
36. Multi-Agent RL: A One-Slider
36
Environment
Observation,
Reward Action
Agent N
Agent 1
.
.
.
Action 1
Action N
Obs. & Reward 1
Obs. & Reward N
39. Knowledge Infused Reinforcement Learning
(KiRL)
39
Knowledge-infused Reinforcement Learning is a class of Neuro-
Symbolic AI techniques that incorporates knowledge (lexical,
domain-specific, common-sense, and constraint-based) into the
policy function, resulting in better performance, interpretable models
and user-level explanations.
44. Overall Execution Flow: Mental Health
44
❏ User Profile Graphs:
❏ Historical Information about
patients stored as Graph
❏ User Profile Graph + Clinical
guidelines drives the high level
decision making of the RL algorithm.
❏ The knowledge modifies the policy to
tend towards the knowledge as a
strong prior in a Bayesian formulation
(We derive a MAP estimate by
formulating it as an optimization
problem)
45. Example: Information Gathering
45
❏ Once the high level task is chosen, it is
expanded to low level execution.
❏ For example what information to gather is
obtained from a dialogue generation
system that takes as input the patient
profile
❏ The responses are parsed for new
information to be added to the patient
knowledge graph.
49. Process Knowledge-based User level explanation
55
Process Knowledge Structure in C-SSRS
I wish I could give a shit about what would
make it to the front page. I have been there
and got nothing. Same as my life. I do have a
gun.’, ’I thought I was talking about it. I am
not on a ledge or something, but I do
have my gun in my lap.’, ’No. I made sure
she got an education and she knows how to
get a job. I also have recently bought her
clothes to make her more attractive. She
has told me she only loves me because I
buy her things.
1. Wish to be dead - Yes
2. Non-specific Active Suicidal
Thoughts - Yes
3. Active Suicidal Ideation with
Some Intent to Act - Yes
4. Label: Suicide Behavior or
Attempt
Agreement with Experts
47%
Process
Knowledge (Ours)
70%
XLNet
Yang et al.
NIPS’19
Gaur and Sheth et al. Internet Computing
Under Review
52. Example: Conversational Information Seeking
What is Conversational Information Seeking?
Conversational Information Seeking (CIS) is an emerging research area within Conversation AI that
attempts to seek information from end-users in order to understand and satisfy user’s needs.
A CIS system can assist clinicians in pre-screening or triaging patients in healthcare.
Here, the agent is the propeller of a conversation with a user
What are Information Seeking Questions (ISQs) ?
ISQs differ from other question type (e.g Clarifying questions, Follow-up questions) by having a
● Structure : semantic relations between questions and logical coherence
● Cover Objective Details
● Expand on the breath of topic
Gaur et al. AAAI 2022
® Samsung Research America
53. Properties of Human's Information Seeking
Questions
60
❏ Task Oriented
❏Seeking information for preparing a delicious cuisine
❏Seeking information about a health condition
❏ Question Styles
❏Contextually, lexically, and syntactically diverse
❏Semantically related and have logical order
❏ Response Shaping
❏Require understanding procedural questions
❏Keep track of entities and actions
Gaur et al. AAAI 2022
® Samsung Research America
54. ISEEQ: Information SEEking Question generation using Dynamic
Meta-Information Retrieval and Knowledge Graphs
61
Gaur et al. AAAI 2022
® Samsung Research America
55. Baseline T5 &
ISEEQ
62
1. Deep Language Models
generate irrelevant questions
(blockchain and currencies) in
absence of external knowledge
2. Without external knowledge,
the model fails to
(a) prevent redundancy
(b) showcase diversity in question
generation
ISEEQ
Baseline T5
Gaur et al. AAAI 2022
® Samsung Research America
56. Example
64
Sentence BERT Encoder
Sentence BERT Encoder
1. What is gross_domestic_product?
2. What is the measure of gross_domestic product?
3. What is the reason nation income relations gross_domestic_product?
4. What is the influence of inflation to gross_domestic_product?
5. What is the meaning of unemployment in inflation?
6. What is the influence of inflation on cost_of_living?
Title: Economy and Employment
Statistics
Description: Learn Information about
key economic concepts including gdp,
inflation, and the influence on
employment
Constituency Parsing
Information + { economy, employment
statistics, employment, influence
employment, inflation influence
employment, gdp, gdp influence
employment, key economic concepts}
economics
economy inflation employment
gdp
gross
domestic
product
unemployment
gnp
gross national
product national
income
cost of living
income
personal
income
income
tax
ConceptNet Graph for Semantic Query Expansion
ISQ by Generative Adversarial Reinforcement Learning
Knowledge-guided
Passage Retriever
Gaur et al. AAAI 2022
® Samsung Research America
Query
57. How ISEEQ show improvement?
(Context: Healthcare)
65
Bothered by feeling
hopeless and
depressed.
Need Advice.
Generator
Network
• Are you feeling bothered?
• Are you depressed?
• Do you feel depressed?
Generated questions are do not
(a) Capture User Context
(b) Share Semantic Relations
(c) Have Logical Order
Qcurr
Baseline Generator-only Approach (T5)
Bothered by feeling
hopeless and
depressed.
Need Advice.
Generator
Network
Neural
Passage
Retriever
Reward
(Qcurr , Qtrue)
• Are you feeling bothered?
• Are you depressed?
• Do you feel depressed?
• Do you feel like you are
depressed sometime?
• Do you know depression can
make you mentally slow?
Generated questions
(a) Captures the context
(b) Share semantic relations
(c) Follow Logical Order
Generated questions are unsafe
Baseline Generator-only Approach (T5) with Neural Passage Retriever
Gaur et al. AAAI 2022
® Samsung Research America
58. 66
How ISEEQ show improvement?
(Context: Healthcare)
Bothered by feeling
hopeless and
depressed.
Need Advice.
Generator
Network
Neural
Passage
Retriever
Context Reward
(Qcurr , Qtrue)
• Do you know what cause depression?
• Can you describe hopelessness?
• Do you think you feel hopeless and
depressed?
• What is the cause of sadness?
Commonsense
Knowledge Graph
ISEEQ Variant 1
Generated questions are
(a) Captures the context
(b) Share semantic relations
(c) In logical Order
Gaur et al. AAAI 2022
® Samsung Research America
59. 67
How ISEEQ show improvement?
(Context: Healthcare)
Bothered by feeling
hopeless and
depressed.
Need Advice.
Generator
Network
Evaluator
Network
ISEEQ Variant 2
Context Reward
(Qcurr , Qtrue)
Order
(Qcurr, Qprev)
Neural
Passage
Retriever
Qcurr: Current Generated Question
Qprev: Previous Generated Question
Qtrue: Ground truth Question
1. Do you think you feel down most of the
time?
2. How often do you feel depressed or
hopeless?
3. How long have you struggled with
depression?
4. Do you know what cause depression?
Generated questions are
(a) Captures the context
(b) Share semantic relations
(c) In logical Order
Gaur et al. AAAI 2022
® Samsung Research America
61. System 1 and System 2 Synchrony
69
Sheth and Thirunarayan, Duality of Data and Knowledge, IEEE Computer Society 2021
Low-level Data
Sensors, Text,
Image, and
Collection
Neural Network
and Deep
Learning
Knowledge (rules,
graphs, process
knowledge, ..)
Symbolic
Reasoning
Decisions/Actions
System 1
System 2
Neural Network
and Deep
Learning
Decisions/Actions
System 1
Low-level Data
Sensors, Text,
Image, and
Collection
Daniel Kahneman - 2011
Thinking Fast and Slow
62. Corpus of
Linguistic
Acceptability
Summarizing
Clinical
Interviews
DSM-5 and PHQ-9
Flesch Reading,
Divergence,
Theme Overlap
Matthew Correlation
Rich
Evaluation
metrics
Stanford
Sentiment
Treebank
Assessing Severity in
User-generated Content
Ordinal Error,
Perceived Risk
Measure, Ranked
Precision/Recall
DSM-5 and Drug
Abuse Ontology
Accuracy
Question NLI
ConceptNet and
WordNet
Concept Mover
Distance, BLEURT
F1-Score and
Accuracy
User-language
Paraphrase
Corpus
Microsoft
Paraphrase
Corpus
Recognizing
Textual
Entailment
Conversational
Information
Seeking
Process Knowledge
NLG
Gaur et al. JMIR’21 Gaur et al. Pone’21,
WWW’19, ACL’ 22
Roy & Gaur et al. ACL
Under Review
ConceptNet,
WikiNews, Wikipedia ,
MS-MARCO
Logical Coherence,
Semantic Relevance,
BLEURT
Accuracy
PHQ-9, GAD-7,
C-SSRS
Accuracy
Avg. # Unsafe
Matches, Avg.
#KG concept
Matches,
Avg. Sq. Rank
Error
Reagle & Gaur
FirstMonday’22
General Language Understanding Evaluation
(GLUE) (Wang et al. ICLR’19)
Knowledge-intensive Language Understanding (KILU)
(Sheth and Gaur IEEE IC’21)
Gaur et al. AAAI’22
Publicly Available Datasets with Knowledge Sources
66. References
74
❏ Kaushik, R., Gaur, M., Zhang, Q., & Sheth, A. (2022). Process Knowledge-infused Learning for Suicidality
Assessment on Social Media. Scholar Commons.
❏ Sheth, Amit, Manas Gaur, Kaushik Roy, and Keyur Faldu. "Knowledge-intensive language understanding for
explainable ai." IEEE Internet Computing 25, 2021.
❏ Manas Gaur, Kalpa Gunaratna, Vijay Srinivasan, and Hongxia Jin. "ISEEQ: Information Seeking Question
Generation using Dynamic Meta-Information Retrieval and Knowledge Graphs." arXiv preprint arXiv:2112.07622
(2021). In AAAI 2022
❏ Kaushik Roy, Qi Zhang, Manas Gaur, and Amit Sheth. "Knowledge Infused Policy Gradients for Adaptive
Pandemic Control." (2021). In AAAI 2021
❏ Kaushik Roy, Qi Zhang, Manas Gaur, and Amit Sheth. "Knowledge infused policy gradients with upper
confidence bound for relational bandits." In ECML-PKDD 2021
❏ Ugur Kursuncu, Manas Gaur, and Amit Sheth. "Knowledge infused learning (k-il): Towards deep incorporation of
knowledge in deep learning." In AAAI 2019
❏ Gaur, Manas, Keyur Faldu, and Amit Sheth. "Semantics of the black-box: Can knowledge graphs help make
deep learning systems more interpretable and explainable?." IEEE Internet Computing, 2021
❏ Sheth, Amit, Manas Gaur, Ugur Kursuncu, and Ruwan Wickramarachchi. "Shades of knowledge-infused learning
for enhancing deep learning." IEEE Internet Computing, 2019
67. Acknowledgement
75
Contribution to Demo on Virtual Assistant for Mental Health
We acknowledge partial support from the National Science Foundation (NSF)
award # 2133842 “EAGER: Advancing Neuro-symbolic AI with Deep Knowledge-
infused Learning,” with PI Dr. Amit Sheth. We also acknowledge partial support
from University of South Carolina ASPIRE Award
Any opinions, findings, and conclusions or recommendations expressed in this
material are those of the author(s) and do not necessarily reflect the views
of the National Science Foundation or University of South Carolina.
Vedant Khandelwal
Ph.D. Student, AIISC
68. Questions?
For further details, please send email to
mgaur@email.sc.edu
More Project on Knowledge Graphs and Knowledge-infused Learning:
http://wiki.aiisc.ai/
76
Editor's Notes
Harmful ---- Hallucinations
Google’s LaMDA
We are focusing on domain-specific issues : https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9195554
One way we can constrain safety by using guidelines
These put a bound on topics covered -- solving patient need and prevent system from going into unsafe turfs
Medical Conditions are ok
Medical conditions that we leave it for human --- constraint our agent to remain in the bounds
Percentage of Questions generated are unsafe
One Example -- Which is unsafe in Vaswani et al.
Which is unsafe in Raffel et al.
You blocked ---- Why KiLSTM model blocked it?
With this process knowledge, the conversational agent can sense when the generated question is safe and when it is unsafe
That information is collapsed into a label which results into information loss
This is under construction
Where I want to go --- come up with a way -- latent representation (low level) -- scale up to abstraction --
Perception -- representation by Neural Network
Cognition -- representation by external knowledge
Knowledge for proper interpretation of the data
Interpretable : Knowledge give you additional guidance, what is the relevance to algorithmic choices
How the data drove the particular pathway
The information you are deriving is not meaningful to outside world
Some relational features that are learned with their English descriptions. It can be seen that the features allow finer grained control at the level of individual shops, homes, residences, workplaces, and routes.
System 1: Processes low level data and enables decisions/actions (This is insufficient as low level data is not at a level of abstraction that humans understand or that can generalize across tasks)
System 1 and 2: System one can process low level data which can be lifted to a higher level of abstraction by mapping to external knowledge. Symbolic reasoning over the mapped knowledge is then performed to enable decisions/actions.
Knowledge Intensive Language Understanding (X axis) vs GLUE (Y axis).
Sensory data abstracted and compiled as a PKG and also processed through a multimodal modeling pipeline. The multimodal model (deep network) interacts with the PKG and provides the assessment of ASD.
Add flowchart, put images on right and text on left. What is Knowledge Infused RL, limitations of current RL.