shallow neural nets as smooth variational problems
a new formulation replaces discrete neural network training with a globally well-posed variational problem over parameter densities, enabling direct solution via a linear system.
aitag research
summaries filed under research
measuring lag relevance in time series forecasts
new model-agnostic method uses shapley values and ghost variables to measure lag importance in univariate time series forecasting.
aiagentic search for federated learning algorithms
auto-fl-research uses coding agents to automatically propose and test federated learning algorithm changes under fixed compute budgets.
aiwiola: a new small language model architecture from scratch
wiola introduces five novel components for efficient small language models, including spiral positional encoding and adaptive token merging.
aipath planning for air traffic control with human-friendly design
a new algorithm generates all safe aircraft paths to help controllers make faster, more transparent decisions.
aigoogle introduces tabfm for zero-shot tabular predictions
tabfm is a foundation model that predicts on new tables without training, using in-context learning and synthetic data.
aimmm data model for knowledge interoperability
a new data model called mmm aims to improve how knowledge is structured, shared, and reused across systems, moving beyond traditional documents.
aihow low-dimensional data structure shapes classification
a mathematical framework extends cover's function-counting theory to analyze how low-dimensional data structure affects binary classification capacity and generalization.
aihumanity's last exam is a distraction
a look at the hle benchmark, why it was created, and the divided expert opinions on its value for evaluating ai systems.
aifiltered mixture-of-generators for synthetic survival data
fogs selects plausible synthetic samples from multiple tabular generators to improve downstream survival model training on small clinical datasets.
aifourier neural operators learn pde solutions efficiently
new theory proves fourier neural operators can approximate and learn time-dependent solutions of dissipative partial differential equations with polynomial sample complexity.
aideep multitask learning for mixed outcomes with shared sparsity
a new method uses deep neural networks and rank-based optimization to handle mixed outcome types in multitask learning with shared predictor selection.
aibounded morality: a computational framework for finite agents
a new framework models moral cognition as a tradeoff between moral breadth and depth under limited resources, recasting ethical theories as efficient strategies.
aisemiscope: tuning gains in semi-supervised security classification
a study disentangles whether improvements in semi-supervised learning for security come from tuning the classifier alone or from joint optimization with the ssl pipeline.
aiai alignment must handle shifting human preferences
a new framework treats alignment as controlling how preferences evolve through interaction, not just satisfying fixed goals.
aigrpo, dr. grpo, and dapo are three settings of one dial
three popular language model training methods all adjust the same number: the standard deviation of correctness marks across sampled answers.
aigoogle ai updates from june 2026
google announced gemini 3.5 live translate, android 17 features, and a new home speaker, plus tools for research and education.
aiwhen feedback actually helps ai improve
a controlled study finds that multi-turn improvement often comes from resampling or extra computation, not from feedback itself, with only strong external teachers providing real gains.
aiwhen does learning to stop help reasoning models
a study shows learned stopping rules can improve reasoning model efficiency on math tasks but not on multiple-choice benchmarks.
aibayesbench tests if llms update beliefs like bayesian reasoners
bayesbench evaluates how large language models update beliefs across multi-turn conversations, comparing their trajectories to rational bayesian inference.
ai