TL;DR
This work proposes a flexible Bayesian framework that significantly outperforms standalone ResNet, human, and LLM-based approaches, demonstrating the practical benefits of fusing these heterogeneous signal sources.
Experimental results demonstrate that SSL outperforms traditional supervised learning in accuracy, scalability, and transferability, while also reducing data labeling costs.
It is shown that a key factor is the error correlation structure between human and AI predictions, and when the AI's prediction errors are negatively correlated with those of the human, the decision maker can construct robust strategies which guarantee improvements in expected utility.
How agents acquire abstract concepts from sparse, diverse examples—often without explicit supervision—remains a central problem in cognitive science and artificial intelligence. Human studies suggest that this ability depends on mental bootstrapping, the gradual construction of complex concepts from simpler partial structures. Building on this idea, we develop a self-supervised framework that trains models on systematically simplified versions of abstract reasoning tasks containing incomplete but structured concept cues. This algorithm enables models to form internal abstractions under limited resources and later apply them to more complex problems. We evaluate the framework across 12 abstract visual reasoning datasets testing in-distribution concept induction, out-of-distribution generalization, and few-shot learning. To contextualize performance, we also measure human accuracy on the same tasks. Models trained on simplified problems generalize robustly, reaching or even surpassing human-level performance. These findings show that abstract reasoning can emerge from structured simplification and minimal data, offering a computational account of concept learning in humans and machines.
Mixture-of-Experts (MoE) models provide a flexible framework for partitioning complex prediction problems into simpler local learning tasks through an input-dependent gating mechanism. Existing interpretable MoE approaches, such as Mixture of Decision Trees (MoDT), achieve transparency by employing homogeneous decision-tree experts, but this restricts the model to a single inductive bias across all regions of the feature space. We extend the MoDT framework by introducing heterogeneous expert families comprising decision trees, linear support vector machines, and quadratic discriminant analysis under a common probabilistic gating mechanism. To ensure coherent likelihood-based inference, non-probabilistic experts are calibrated to produce conditional class probabilities, allowing parameter estimation within the generalized Expectation-Maximization framework of MoDT. We further establish theoretical monotone ascent guarantees for the proposed heterogeneous gating updates, providing a justification for the optimization procedure. Experiments on a diverse collection of synthetic and real-world benchmark datasets demonstrate that the proposed framework adaptively specializes experts according to local data geometry, yielding interpretable expert assignments while achieving predictive performance competitive with homogeneous MoDT and Random Forests. The proposed approach combines interpretability, adaptive inductive bias selection, and probabilistic coherence within a unified mixture-of-experts framework.
This work introduces Co-RL, a framework in which multiple decoupled models, sharing no parameters, are simultaneously optimized through RL using rewards derived from their peers, and shows that unsupervised reasoning can emerge through cooperative multi-agent training.