BIRDNet is a neural network that mines Boolean implication relationships (BIRs) from tabular data and encodes them as a sparse, interpretable architecture where each hidden unit represents a mined rule connecting two features. The design yields a model that is at most 2/d of the weights active per layer and retains symbolic identities for each unit, allowing direct rule extraction without surrogate models. Experiments on six transcriptomic and proteomic datasets show BIRDNet achieves AUROC within 0.02 of the best dense baseline while using up to 95× fewer active parameters, and its first‑layer rules align with known biological signatures.
By Tirtharaj Dash
arXiv:2605.06240v2 Announce Type: replace-cross
Abstract: Forward-Forward (FF) training lets each layer learn from a local goodness criterion. In cumulative-goodness variants, later layers can inheri...
By Amirhossein Yousefiramandi
The paper investigates four ways to grow a classifier—adding a tree level, a hidden unit, a leaf split, and a statistically significant split—under a fixed protocol for tree‑structured and constructive models. It shows that the most natural method of deepening a soft decision tree by duplicating a leaf’s class distribution leaves the gradient of new gates identically zero, preventing learning, and proposes a small random perturbation as a fix. The other three growth decisions each provide a distinct benefit: fitting a new hidden unit to residual error yields a smaller network, splitting the leaf with the largest expected error adds sparsity, and requiring statistical significance before splitting adds no value and reduces accuracy.
By Cagri Temel
arXiv:2606. 17886v1 Announce Type: new Abstract: Monotonicity has been a long-running architectural inductive bias for neural networks, motivated by tabular, scientific, and economic settings where outputs are known to respond monotonically to certain inputs.
By Mikhail Krasnov, Carolina Fortuna, Bla\v{z} Bertalani\v{c}
arXiv:2606. 10632v1 Announce Type: cross Abstract: Lipschitz-style individual fairness formalizes the idea that semantically similar examples should receive similar predictions, but its evaluation in multi-task learning (MTL) can be confounded by method-induced representation scales.
By Junbo Ding, Xin Zang, Chenchen Pan, Donghao Song, Jiaxin Zhu, Danhuai Guo
arXiv:2512. 22240v5 Announce Type: replace-cross Abstract: Machine learning models are primarily judged by predictive performance, especially in applied genomics, where explanations are read as biological findings.
By Chama Bensmail
Monotonicity has been a long-running architectural inductive bias for neural networks, motivated by tabular, scientific, and economic settings where outputs are known to respond monotonically to certain inputs. Existing approaches are MLP- or flow-based and lack per-edge functional transparency; the only Kolmogorov--Arnold Network (KAN) variant with monotonicity, MonoKAN, enforces the constraint only on a restricted parameter subset and requires a projection-style training procedure.
arXiv:2607. 04595v1 Announce Type: new Abstract: Most classification problems assume the classes are roughly separable, so that an individual sample can usually be assigned to one class.
By Youssef Marrakchi, Davide D'Ascenzo, Sebastiano Cultrera di Montesano
RoBell-RVFL is a lightweight, quality‑aware generalized bell random vector functional link network designed to address class imbalance and noisy data in real‑world datasets. It uses a dual‑strategy sample‑level weighting: unit weights preserve minority class information, while a probability‑weighted generalized bell membership function suppresses noisy majority samples in a kernel‑induced feature space. Experiments on UCI and KEEL benchmarks, including tests with up to 40% label noise, show that RoBell‑RVFL consistently outperforms recent RVFL variants, demonstrating the importance of adaptive, quality‑aware sample weighting for robust learning.
By A. Rahaman, A. Quadir, M. Tanveer
arXiv:2602. 01083v2 Announce Type: replace Abstract: Weight-space learning studies neural architectures that operate directly on the parameters of other neural networks.
By Adir Dayan, Yam Eitan, Haggai Maron
The paper investigates two strategies for improving large language model (LLM) evaluation: specialized judge weights and rule‑based deferral policies. Experiments on nearly 100,000 rubric‑conditioned samples show that correct rubrics boost accuracy, while incorrect ones hurt it, and that splitting training data into criterion‑specific experts can severely degrade performance unless the experts are warm‑started from a unified model. The authors demonstrate that lightweight deferral cascades can match or exceed the accuracy of larger standalone judges at a fraction of the compute cost, and they provide practical design rules for building efficient, reliable LLM evaluators.
By Ye Chen, Weining Zhang
arXiv:2602. 22422v2 Announce Type: replace-cross Abstract: Smooth-basis models such as Chebyshev polynomial regressors and radial basis function (RBF) networks are well established in numerical analysis.
By Luciano Gerber, Huw Lloyd