Jay (Jianqiang) Wang’s Post

#A #Guide #to #Investigating #the #Datafied #State #Through #Documents #JennaBurrell #Professor #UCBerkeley #Director #Data & #SocietyResearchInstitute "As a way to redirect attention to #who #holds #power, and to understand #how #they #wield #it, it is worth our time to treat #document #analysis #as #a #method. #Where #research #calls #the #accounts #of #those #in #power #into #question, #it #helps #to #come #with #receipts."

📝 Announcing our new paper that proposes a framework to enhance causal reasoning and explainability in LLMs 🔹 "Cause and Effect: Can Large Language Models Truly Understand Causality?" 🔹 In collaboration with Carnegie Mellon University, University of North Texas, Rensselaer Polytechnic Institute, and University of Massachusetts Amherst 🔹 We introduce the Context-Aware Reasoning Enhancement with Counterfactual Analysis (CARE-CA) framework to enhance causal reasoning and explainability in LLMs. It integrates explicit causal detection using ConceptNet and counterfactual statements, alongside implicit causal detection through LLMs, and introduces CausalNet, a new dataset for further research in causal reasoning. 🔹 PDF: https://lnkd.in/gcXuFBgH ✍🏼 Authors: Swagata Ashwani, Kshiteesh Hegde, Nishith Reddy Mannuru, Mayank Jindal, Dushyant Singh Sengar, Krishna Chaitanya Rao, Dishant Banga, Vinija Jain, Aman Chadha #artificialintelligence #research

Great news 🎉 Congrats S.Siamak Ghodsi for the accepted paper on fair graph clustering at PAKDD24. While there are a lot of works on tabular data fairness, graph data fairness has been less explored. In this work, we adopt the idea of individual graph fairness, i.e., the proportion of a node's 1-hop neighbors in its own cluster should be similar to the proportion of 1-hop neighbors belonging to the entire population. For details on the motivation, method and the link to the pre-print, see Siamak's post. The work is funded by the ITN NoBIAS project and is also related to multi-modal fairness addressed in the MAMMOth project. #responsibleAI, #fairness #graphfairness #noniidfairness #fairgraphclustering

New year, new blog post: https://lnkd.in/dadWHrZa . Taking a step back to think about two approaches to managing academic quality and standards (quant data; periodic review) and how they can (or perhaps should) dovetail together.

A cracking thought piece to kickstart the New Year: here’s a quote from it that will appear (cited) in a forthcoming publication from me for sure: Education is a social process. When we’re dealing with social processes quantitative data can indicate something you probably need to look at and think about, but it frequently can’t explain the thing that it’s highlighted. That requires other, more qualitative approaches together with the application of expertise and practical wisdom.

An interesting kick off towards self-optimization of LLMs. Exciting times! Just make sure to notice the limitations section at the end ofthe paper to fully comprehend the scope 👾

A complete overview of LLM Agent Applications! 🙏 This is a refreshing survey paper, that goes beyond simply documenting the current landscape. It starts by giving the reasoning behind different approaches and describes how these are implemented by various papers in a very intuitive way. It divides the Large Language Model requirement part into three: - Brain: Acting as the reasoning engine, memory and long term planning - Perception: This section covers all modalities possible - Action: Using Text, External Tools and acting independently Finally the paper covers implementation in single agent, multiple agent and human-agent settings. The paper is really comprehensive and actually reads like a very upto date book on the topic. An amazing 50 page read for your Saturday: https://lnkd.in/dzpQdkDs

Here is a seminar I gave recently, called "Open Science, reproducible research, and the citation of articles, code and data alike". After touching upon the principles of Open Science and reproducible research, I focus on practical approaches to make research software and data Findable, Accessible, Referenceable and Citable (FARC), as well as good practices to cite others' code and data. I finally make the point that to go further, editors must now actively support code and data citation, evaluation metrics must change to give credit for and reward impactful code and data, and all of that shall require a true cultural shift of researchers' mindset. https://lnkd.in/dZZfmTfp

The original article was posted last week, but this is a good follow up by Pascal Biese. The idea that structured context is less efficient through an LLM should really surprise no one. Structure (semantification) costs energy to produce, especially when the primary training data for these LLMs is primarily narrative and conversational. This requires more computational cycles, requires more inference from existing samples, and requires more validation. This is a key argument for knowledge graphs. The semanticification has (largely) been done with a knowledge graph. These means that such structured content is the equivalent of a high carb meal.