KL- Data Science - HealthCare- TrustifAI

[xp1632]

https://jobs.dfki.de/intern/ausschreibung/data-science-researcher-554367.html

The part I can contribute to based on their description:

The TrustifAI project aims to contribute a set of concrete solutions to improve the trustworthiness of AI applications in health and wellbeing at different stages of the development lifecycle. A quality platform for developing trustworthy AI applications will enable users to build efficient and effective data science analysis pipelines through a human-in-the-loop approach, with the aim of increasing trustworthiness.

---

Contact Person: David Antony Selby Group: https://datasciapps.de/ Professor: Sebastian Vollmer (their publication is quite good: David, Sebastian)

Their project of hiring PhD: https://datasciapps.de/job/curatime/

Phd salary: https://datasciapps.de/phd/ A PhD scholarship typically comprises a monthly stipend of approximately €1700, supplemented by a €600 HiWi (research assistant) position.

File submission place: https://vollmer.kl.dfki.de/

---

• My interests in this position are :

  • I am interested in Healthcare and trustworthy AI

• My strength is my project experience and research ability

• My shortcoming is only have basic experience in Machine Learning

  • where i am taking trustworthy AI course and reviewing pytorch (for application of machine learning part)

  • and algorithmic data analysis (for the complex unstructured data in Health data )

• My thoughts about this project:

• Is my insight of this project:

• The key points in Trustworthy AI in healthcare are:

  • its explainability , because trust built on understanding
  • its accuracy, unlike prediction in other fields, the robust and accuracy are crucial for it is related to a real life
  • its ability to process complex data in different ways (DNA, longitude data, etc) and give a unified system to weight their impact

---

For the explainablity part, I am thinking about

Retrieval-augmented generation (RAG) is a technique for enhancing the accuracy and reliability of generative AI models with facts fetched from external sources. https://blogs.nvidia.com/blog/what-is-retrieval-augmented-generation/ https://www.promptingguide.ai/techniques/rag

• I am not sure if this is the perfect match, because this is mostly used in NLP field that offer reference of source for chatgpt modals
• but I am sure a part of the distrust comes from not knowing the source or evidence of a certain diagnose
• by providing the source or explain the reason of diagnosing could improve the persuasion of an AI health care diagnose

---

!important While writing this, I think about we let the patient to ask questions or (prompts) to their diagnoses, so they could better understand,

---

Here are some references of using RAG (Retrieval-augmented generation) by Langchain (https://smith.langchain.com/o/2964ca36-6631-5a87-98be-21e79c33ca70/):

Video:

1. Local Retrieval Augmented Generation (RAG) from Scratch (step by step tutorial) https://www.youtube.com/watch?v=qN_2fnOPY-M (long from scratch) git: https://github.com/mrdbourke/simple-local-rag

2. RAG + Langchain Python Project: Easy AI/Chat For Your Docs https://www.youtube.com/watch?v=tcqEUSNCn8I (short and concise)

---

Text

1. Retrieval-augmented generation, step by step https://www.infoworld.com/article/3712860/retrieval-augmented-generation-step-by-step.html

2. What's Langchain and example: https://www.infoworld.com/article/3706289/what-is-langchain-easier-development-of-llm-applcations.html

PyTorch All In guide: Learn PyTorch for deep learning in a day. Literally. https://github.com/mrdbourke/simple-local-rag

---

Paper:

These are the papers I read about trustworthy AI in healthcare:

1. Trustworthy Artificial Intelligence in Healthcare https://www.theseus.fi/bitstream/handle/10024/785071/KhanUAlamakiATrustworthyArtificialIntelligenceInHealthcare.pdf?sequence=1