Université Paris-Saclay

In a Mesa implementation of the Schelling segregation model, one agent's symbolic rule for classifying neighbors is replaced by a locally served Qwen3.5 model that classifies through tool calls, while the rest of the dynamics stays fixed. Statistical model checking with MultiVeStA shows the expected ratio of happy agents rising from about 0.65 to about 1.0 within 60 steps in both variants, with point-wise t-tests finding no difference, yet the LLM version takes 8,917 seconds against about 2 seconds. The 0.8b and 2b models show semantic errors, wrong tool-call counts or stuck generation in unit-style checks, while 4b and 9b pass all of them.

In a Mesa implementation of the Schelling segregation model, one agent's symbolic rule for classifying neighbors is replaced by a locally served Qwen3.5 model that classifies through tool calls, while the rest of the dynamics stays fixed. Statistical model checking with MultiVeStA shows the expected ratio of happy agents rising from about 0.65 to about 1.0 within 60 steps in both variants, with point-wise t-tests finding no difference, yet the LLM version takes 8,917 seconds against about 2 seconds. The 0.8b and 2b models show semantic errors, wrong tool-call counts or stuck generation in unit-style checks, while 4b and 9b pass all of them.

The work shows that semi-supervised domain generalization methods such as FBCSA and DGWM degrade substantially under long-tailed class distributions, and proposes IMaX: maximizing the mutual information between learned features and latent labels under supervision constraints from labeled samples, while replacing the standard marginal entropy term with a Tsallis α-entropy to relax the class-balance assumption; across two modalities (ESCA histopathology and diabetic retinopathy grading), three SSL frameworks and three SSDG methods, IMaX improves accuracy in all but one setting, with gains up to 7.3 points in the low-label regime (mL=5).

The work shows that semi-supervised domain generalization methods such as FBCSA and DGWM degrade substantially under long-tailed class distributions, and proposes IMaX: maximizing the mutual information between learned features and latent labels under supervision constraints from labeled samples, while replacing the standard marginal entropy term with a Tsallis α-entropy to relax the class-balance assumption; across two modalities (ESCA histopathology and diabetic retinopathy grading), three SSL frameworks and three SSDG methods, IMaX improves accuracy in all but one setting, with gains up to 7.3 points in the low-label regime (mL=5).

This work developed and evaluated a fully on-premise clinical AI agent that couples local operational control with a multi-perspective reliability framework to support selective autonomy, achieving 90.04% accuracy on a seven-disease task and 83.8% on a four-disease task across two MIMIC-IV-derived benchmarks, and finding that diagnostic behavioral consistency best discriminated correctness (AUC = 0.860, and AUC = 0.875 under stress testing), with a consistency threshold of 0.90 retaining 49.4% of cases at 98.9% diagnostic accuracy.

This work developed and evaluated a fully on-premise clinical AI agent that couples local operational control with a multi-perspective reliability framework to support selective autonomy, achieving 90.04% accuracy on a seven-disease task and 83.8% on a four-disease task across two MIMIC-IV-derived benchmarks, and finding that diagnostic behavioral consistency best discriminated correctness (AUC = 0.860, and AUC = 0.875 under stress testing), with a consistency threshold of 0.90 retaining 49.4% of cases at 98.9% diagnostic accuracy.