探索全球前沿学术脉络

GeroQubit: a lightweight, honesty-first de-novo design platform for geroscience-native small molecules with calibrated uncertainty

Computational molecule generation has outpaced its own credibility. We present GeroQubit, a GPU-free de-novo design platform that organizes candidates along a target x tissue x hallmark model and reports every signal alongside its measured baseline. We treat our tissue aging-signature readout as a mechanistic structural prior that we explicitly disclose is not validated against lifespan, and we surface efficacy only through a structure-to-lifespan k-NN whose weak but real signal (leave-one-out rho ~ 0.145) is wrapped in empirically-calibrated conformal intervals (90% target, 90.3% measured coverage). On a held-out retrospective recovery of ~1,940 ChEMBL binders against decoys, the score reaches ROC-AUC 0.945 with ~20x enrichment at 1% (BEDROC 0.91) and survives a scaffold-disjoint split - yet we report that it collapses to near-random (AUC 0.62) on genuinely novel chemotypes. Molecules are assembled reaction-first, so every candidate carries a verified synthetic route and atom-level synthon provenance; ADMET is handled as a multi-objective Pareto problem. We frame the disclosed weak signals and the hard-case failures not as flaws but as the honest, decision-useful output the field's own critics demand.

Bias-Controlled Primal-Dual Natural Actor-Critic: Optimal Rates for Constrained Multi-Objective Average-Reward RL

arXiv:2606.25012v1 Announce Type: new Abstract: Many reinforcement learning (RL) problems in the infinite-horizon average-reward setting require optimizing multiple conflicting objectives while satisfying multiple safety constraints. A common approach is concave scalarization, where the agent maximizes a utility $ f(J^\pi_{r_1}, \ldots, J^\pi_{r_M}) $ subject to a scalarized constraint $ g(J^\pi_{c_1}, \ldots, J^\pi_{c_N}) \ge 0 $, where $J^\pi_{r_m}$ and $J^\pi_{c_n}$ denote the average-reward and cost under policy $\pi$. However, the nonlinearity of $f$ and $g$ introduces bias in policy-gradient and actor-critic methods, since gradients must be evaluated using noisy estimates of $J^\pi,$ and $ \mathbb{E}[\partial f(J^\pi)] \neq \partial f(\mathbb{E}[J^\pi]),$ and this bias propagates through both primal and dual updates. We propose an MLMC-based primal-dual Natural Actor-Critic algorithm for average-reward MDPs that controls bias in scalarized objectives, constraint evaluation, and actor-critic estimation without requiring mixing-time knowledge. We show that the algorithm achieves optimal global convergence and constraint-violation rates of $ \tilde{O}(1/\sqrt{T}) $. To our knowledge, this is the first result establishing optimal convergence for concave scalarized multi-objective RL in the average-reward setting, both with and without constraints, and the first to do so without mixing-time information even in the absence of scalarization.