A novel learned optimization framework improves the stability and convergence of solutions for ill-posed inverse problems, offering a significant advancement in fields like brain imaging.
A new analysis reveals the core limitations of graph neural networks and introduces strategies to overcome performance bottlenecks in complex graph data.
New research reveals that subtly manipulating the explanations accompanying AI decisions – even correct ones – can significantly erode human trust in these systems.
New research explores a method for anticipating problematic behaviors in large language models by analyzing the data they’re trained on, rather than assessing the models themselves.
Researchers have developed a novel deep learning method that leverages advanced state space models to identify subtle anomalies in hyperspectral imagery.
![The projection of model capabilities-ranging from baseline performance [latex] (orange curves) [/latex] to enhanced reasoning with an advanced base model [latex] (gpt-5.1-codex-max, green curve) [/latex]-reveals an overall trajectory [latex] (blue curve) [/latex] predictably limited by the estimated 50% model horizon as defined by METR.](https://arxiv.org/html/2602.04836v1/fig/extrapolation_spline.png)
A new analysis challenges the prevailing assumption of continually accelerating AI capabilities, suggesting diminishing returns may be on the horizon.
A new method illuminates the structural changes needed to alter predictions made by hypergraph neural networks, offering a crucial step towards trustworthy AI.
![The Deep Rational-ANOVA Network (RAN) employs a deep residual backbone facilitating sparse pairwise interactions and node-wise updates, coupled with learnable rational units-initialized for identity and stabilized with positive denominators [latex]1 + \text{softplus}(\cdot)[/latex]-to enable pole-free composition and robust signal processing.](https://arxiv.org/html/2602.04006v1/x2.png)
A new deep learning architecture leverages mathematical decomposition to create more stable, interpretable, and powerful models.
![Despite achieving high predictive accuracy on observed neural activity, unconstrained model exploration failed to uncover the underlying mechanisms of a simulated neuromechanical system, while a structure-constrained search - guided by prior knowledge of potential connections - successfully identified the correct signs and magnitudes of interactions within the neural circuit, demonstrating that mechanistic discovery requires incorporating structural priors beyond mere predictive performance-even when working with [latex]in\thinspace silico[/latex] testbeds.](https://arxiv.org/html/2602.04492v1/x1.png)
New research leverages detailed simulations and connectome data to uncover the underlying mechanisms of neural circuits.
![Machine learning frameworks exhibit vulnerabilities stemming from diverse sources, including adversarial attacks targeting model robustness - often formulated as finding perturbations δ such that [latex] f(x + \delta) \neq f(x) [/latex] - and data poisoning which compromises training data integrity, alongside issues related to model privacy and security concerning sensitive information leakage or manipulation.](https://arxiv.org/html/2602.04753v1/img/study1codebook1.png)
New research reveals a disconnect between how adversarial machine learning threats are understood and addressed in industry versus academic settings.