courses:xai:p1_2025

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision		 Previous revision
courses:xai:p1_2025 [2025/11/20 10:14] – [Eyetracking for analysis of XAI and human behaviour in anomaly-deteciton tasks] admincourses:xai:p1_2025 [2025/12/01 09:24] (current) – [Programming Assignment I] admin
Line 3: Line 3:
 Assignments marked with {{:courses:xai:winner.png?25|}} can be extended as a continution in Programming Assgnment II. Assignments marked with {{:courses:xai:winner.png?25|}} can be extended as a continution in Programming Assgnment II.
 Projects marked with {{:courses:xai:contract.png?25|}} can be (with some additional work and depending on the results) published as scientific papers. Projects marked with {{:courses:xai:contract.png?25|}} can be (with some additional work and depending on the results) published as scientific papers.
 +
 +Select your project: 
 +  * [Monday group]: [[https://ujchmura-my.sharepoint.com/:x:/g/personal/szymon_bobek_uj_edu_pl/IQDwhgwVco6YS53jp8DRACWHAcdSSHvoObQeqkbFPXpTwHc?e=YSBAr2| Choose your project]]
  
  
Line 31: Line 34:
  
 Possible extensions to a paper/master thesis: how to use thes emethod to other than tabular cases/models, for instance: Graph neural networks, or images? Possible extensions to a paper/master thesis: how to use thes emethod to other than tabular cases/models, for instance: Graph neural networks, or images?
 +
 +
 +===== WinClip for explainable anomaly detection=====
 +{{:courses:xai:winner.png?30|}}
 +{{:courses:xai:contract.png?30|}}
 +
 +Testing WinCLIP model for explainable anomaly detection. See if the model can be used to obtain textual explanations of a detrected anomalies. 
 +
 +
 +===== Predictive multiplicity estimation=====
 +{{:courses:xai:winner.png?30|}}
 +{{:courses:xai:contract.png?30|}}
 +
 +Find ready to use frameworks that allwos to measure [[https://arxiv.org/pdf/1909.06677|predictive multiplicity]] -- how much two or more models that provide the same accuracy in predicion are able to provide conflicting predictions in areas not covered in trainig set.
 +
 +You may start by lookint this paper and its references: [[https://proceedings.neurips.cc/paper_files/paper/2024/file/dbd07478c4aac41c0ce411e12f2e5a28-Paper-Conference.pdf|RashomonGB: Analyzing the Rashomon Effect and Mitigating Predictive Multiplicity in Gradient Boosting]]
 +
 +
 +
 +===== Logical neural networks =====
 +{{:courses:xai:winner.png?30|}}
 +{{:courses:xai:contract.png?30|}}
 +
 +How to use [[https://ibm.github.io/LNN/usage.html#using-the-lnn|LNN]] to obtaion explanations of their predictions?
 +For instance how to extract the activated predicates to outrput the rule that resulted in final decission?
 +
 +
 +===== ProtoTS Net =====
 +
 +{{:courses:xai:winner.png?30|}}
 +{{:courses:xai:contract.png?30|}}
 +
 +[[https://github.com/bmalkus/ProtoTSNet/tree/manual-protos|ProtoTSNet]] is a inherently interpretable DNN for time series classificaiton.
 +Inspect its variant which goal is to guide training procedure with manually created prortypes.
 +
 +What could be done to improve its predictive power?
 +
 +
 +===== TSProto for images=====
 +{{:courses:xai:winner.png?30|}}
 +{{:courses:xai:contract.png?30|}}
 +Try to implement TSProto variant that will work for images too (use segmentation to detect segments), cluster segments to detect prototypes, build a decision tree that explains a decision using this visual prototypes.
 +
 +
 +
 +
 +
 +===== Counterfactual Analysis =====
 +{{:courses:xai:winner.png?30|}}
 +
 +Benchmark 3 models on 3 datasets (e.g., Breast Cancer, Adult, Heart Disease), explain the best model using (SHAP, Lime, and feature importance), and get counterfactual explanations using one of these frameworks (ACFX, DICE, CLUE, CFNOW) .
 +
 +**Extra**: Evaluate the counterfactual explanations.
 +
 +===== Human-Constrained Counterfactuals =====
 +{{:courses:xai:winner.png?30|}}
 +
 +Train a classifier on a tabular dataset (e.g., Breast Cancer, Adult, Heart Disease), generate counterfactual explanations for selected instances without any constraints using (ACFX, DICE, CLUE, CFNOW) , then regenerate counterfactuals while forbidding changes to sensitive or immutable features (e.g., age, gender, race, marital status, education). Compare both sets of explanations in a before/after table. Describe how the constraints impact the plausibility, fairness, and actionability of the results.
 +Example: “A bank cannot change a customer’s age or marital status, but can suggest financial improvements.”
 +
 +**Extra**: Quantify how many counterfactuals become invalid or infeasible when constraints are applied.
 +
 +===== Correlation Problems in XAI =====
 +
 +Train a model on a dataset with correlated features (e.g., Breast Cancer, Adult, Heart Disease). Visualize the correlation structure, then generate explanations using two XAI methods (e.g., SHAP and LIME). Identify at least one case where the model attributes importance to a feature that is likely acting as a proxy for another correlated variable. Show before/after visualizations and briefly explain whether the explanation is genuinely meaningful or simply an artifact of feature correlation.
 +
 +**Extra**: Propose and implement one strategy to reduce misleading explanations (e.g., feature grouping, PCA, domain-level feature merging…).
 +
 +
 +===== TSProto for Colab =====
 +
 +Try to run tutorials of [[https://tsproto.readthedocs.io/en/latest/|TSProto]] in Colab (it will require some code tweeking/requirements adjustment to work with numpy 2.0.0.
 +Once done, create a pull request to include the changes in the current implementation of TSProto
 +
 +
 +
 +
 +
  
  
  
  
  • courses/xai/p1_2025.1763633699.txt.gz
  • Last modified: 6 months ago
  • by admin