Omnixai

1. Add FFT preconditioning for feature visualization
2. Implement ScoreCAM, LayerCAM, SmoothGrad and Guided Backpropagation
3. Fix some small bugs.

1. Support feature visualization (an optimization-based method) for vision models.
2. Allow visualizing feature maps in CNN models.
3. Add more tutorials on the supported explainers.
4. Fix some bugs in the ranking explainers.

1. Fix a bug in the MACE refinement module.
2. Add several explainers for ranking tasks, e.g., ValidityRankingExplainer, PermutationRankingExplainer, MACEExplainer.
3. Add save and load functions for the supported explainers.
4. Add a RL-based approach for the MACE counterfactual explainer, e.g., set `method = "rl"` when creating a MACE explainer.

1. Allow LIME and SHAP to compute feature importance scores for a subset of features.
2. Resolve an OOM issue when integrated-gradient is applied on large pretrain language models.
3. Add a new interface for multimodal models.
4. Add explainers for vision language tasks, e.g., GradCAM and integrated-gradient.

1. Re-designed the Timeseries data class.
2. Re-implemented SHAP, CE and MACE for the new Timeseries data class.
3. Fixed some small issues in the dashboard.

1. Support accumulated local effects (ALE).
2. Revise the visualization of PDP results.
3. Fixed some minor issues.