CausalShift: A Modular, Plugin-Based Framework for Dataset Shift Handling in Machine Learning

Seeing Through the Shift: Causality-Inspired Robust Generalized Category Discovery

Learning the Pareto Frontier of Predictive Models under Distribution Shift

Modern machine learning pipelines increasingly rely on reusing pretrained and foundation models across downstream tasks. These pretrained models can differ not only in performance but also in how they can be used: some only provide black-box predictions, while others may permit white-box access to internal representations that can be probed or fine-tuned. When deployed to the target domain in the presence of distribution shift, no single strategy, including zero-shot application, fine-tuning, or directly training a target-specific model, is uniformly the best. In this work, we propose Frontier Learning, a framework that treats a library of candidate models spanning different training histories and access regimes as complementary sources of information rather than mutually exclusive alternatives. Frontier Learning constructs a unified target-domain feature by concatenating internal representations from white-box candidates as well as prediction outputs from black-box candidates, then fits a lightweight, regularized supervised learner on this concatenated representation using labeled target data. Because the resulting hypothesis class contains predictors obtained by zero-shot reuse, fine-tuning, and direct training as special cases, empirical risk minimization over the frontier learner is guaranteed to be no worse, on the training sample, than any individual baseline. We evaluate the framework in simulations spanning varying degrees of source-target compatibility and in two real-world distribution-shift settings: visual domain adaptation on DomainNet/VisDA and clinical mortality prediction across intensive care unit domains using MIMIC-IV-Notes. Across all settings, Frontier Learning matches or outperforms the strongest individual reuse strategy, with the largest gains arising precisely when no single baseline is reliable across the range of shift considered.

A Graph-Based Reinforcement Learning Framework for Structured Drift Diagnosis and Recovery in Autonomous LLM Agents

Autonomous LLM agents are increasingly deployed in complex real-world workflows, yet they remain vulnerable to runtime behavioral drift, a silent deviation from the original task that can lead to irreversible side effects on external systems. Existing approaches address drift at the prompt level but lack structured mechanisms for step-level detection, risk assessment, and recovery decision. Because the main task-executing agent is often a large and expensive model that cannot be re-trained on every deployment, this work targets a plug-and-play recovery module instead. It introduces a graph-based framework in which a single small language model is trained via reinforcement learning to specialize at each node of a recovery graph, external to the main agent. Each node has a precise role\,: drift classification, operation detection, risk evaluation, or final decision and the model learns to produce structured XML-formatted reasoning adapted to that role. Training combines rule-based structural rewards with an LLM-as-judge semantic-quality signal, so that the model is graded both on how it answers (schema and length) and on what it says. Experiments on the public AppWorld benchmark show that the method generally exploits information about the suspected drift onset to issue correct recovery decisions using a small language model. In addition, the trained small language model reliably respects the prescribed output schema and produces semantically appropriate content in each field according to its assigned node role.

Decomposing decisions: causal concept explanations for deep learning models

ABSTRACT The deployment of deep learning models in high-stakes applications such as autonomous driving is critically important, yet their black-box nature remains a fundamental barrier to trust and accountability. Existing explainability methods typically produce ambiguous, pixel-based heatmaps that capture correlation rather than establishing a causal link between high-level, human-interpretable concepts and model outputs. This paper introduces the Causal Concept Decomposer (CCD), a three-stage framework for concept-driven causal explanation of object detectors. CCD first employs semantic segmentation to isolate the target object, then applies Non-negative Matrix Factorization to discover constituent semantic parts, and finally uses Sobol sensitivity analysis to quantify the causal influence of each part on the detector’s decision. Evaluated on the MS COCO dataset, CCD produces explanations that are both visually coherent and quantitatively more faithful than existing approaches, achieving a Deletion AUC of 0.11 and an Insertion AUC of 0.91. By moving beyond correlational attribution towards principled causal analysis, this work represents an important step towards more trustworthy and reliable AI systems.

From Training to Deployment: Post-Hoc Causal Feature Identification via Sensitivity Ratios

The Normalised Sensitivity Ratio (NSR) is introduced, a post-hoc, model-agnostic diagnostic for this question under a structured-shift regime: environments differ primarily in the mean of spurious features while the causal mechanism and causal marginals remain stable, as in multi-site clinical data or multi-batch genomics.

ELT Pipelines Enhanced with Causal Intelligence for Reliable and Bias-Resilient AI Decision Systems

In the world of modern analytics and machine learning, pipelines known as Extract-Load-Transform (ELT) pipelines are commonly used to transform raw data into representations that are suitable for models. But most of the existing ELT workflows are primarily correlation driven and don’t explicitly maintain causality during data transformation. This means that joins, filtering, aggregation and feature preparation steps can create confounding effects, selection bias, or false treatment-outcome relationship that can create unreliable AI-based decisions. This paper proposes such a causal intelligence enhanced ELT framework, where causal reasoning is integrated into the ELT process directly. Three pipeline-native components are proposed: causal graph-constrained data integration via Causal Join, covariate balancing and adjustment in-pipeline via Propensity Balance, and execution optimization via task scheduling and causal artifact reuse via Causal Scheduler. The proposed method is reduces bias earlier in the data lifecycle compared with purely post-hoc causal modelling less biased than post hoc causal modelling approaches as the transformation process itself is less biased. On experiments performed on the IHDP, Twins and Criteo uplift datasets, we have seen improved estimation of treatment effects, lower bias and higher predictive reliability. The proposed framework reduces the error in the estimation by approximately 55-65%, improves predictive accuracy from 82.0% to 91.0%, corresponding to a 9 percentage-point improvement. and gives a $7.5 \times$ speedup performance for GPU compared to CPU. These findings point towards the promise of causal intelligence in making more reliable and efficient decisions within ELT pipelines, potentially enhancing the overall reliability and trustworthiness of AI-driven decision systems in the field.