10 papers
Support Before Frequency in Discrete Diffusion
Adrian Müller, Antoine Gonon, Zebang Shen +2
Discrete diffusion models are increasingly competitive for language modeling, yet it remains unclear how their denoising objectives organize learning. Although these objectives tar…
Revisiting DAgger in the Era of LLM-Agents
Changhao Li, Rushi Qiang, Jiawei Huang +4
Long-horizon LM agents learn from multi-turn interaction, where a single early mistake can alter the subsequent state distribution and derail the whole trajectory. Existing recipes…
Muown: Row-Norm Control for Muon Optimization
Kai Lion, Florian Hübler, Bingcong Li +2
Muon has emerged as a strong competitor to AdamW for language model pre-training, yet its behavior at scale is sensitive to weight decay. Recent work has observed that, for Muon wi…
Select-then-differentiate: Solving Bilevel Optimization with Manifold Lower-level Solution Sets
Saeed Masiha, Zebang Shen, Negar Kiyavash +1
We study optimistic bilevel optimization when the lower-level problem has a non-isolated manifold of minimizers. In this setting, the hyper-objective may be non-differentiable beca…
On the Connectedness of Sublevel Sets in Invex Optimization
Vinzenz Thoma, Zebang Shen, Niao He
Understanding the topology of sublevel sets yields crucial insights into the optimization landscape of non-convex functions. If sublevel sets are connected, local search algorithms…
Manifold Generalization Provably Proceeds Memorization in Diffusion Models
Zebang Shen, Ya-Ping Hsieh, Niao He
Diffusion models often generate novel samples even when the learned score is only \emph{coarse} -- a phenomenon not accounted for by the standard view of diffusion training as dens…
A Schrödinger Eigenfunction Method for Long-Horizon Stochastic Optimal Control
Louis Claeys, Artur Goldman, Zebang Shen +1
High-dimensional stochastic optimal control (SOC) becomes harder with longer planning horizons: existing methods scale linearly in the horizon , with performance often deteriora…
Superquantile-Gibbs Relaxation for Minima-selection in Bilevel Optimization
Saeed Masiha, Zebang Shen, Negar Kiyavash +1
Bilevel optimization (BLO) becomes more challenging when the lower-level objective admits multiple minimizers. Compared with the commonly studied unique-minimizer setting, this int…
Optimal Local Convergence Rates of Stochastic First-Order Methods under Local -PL
Saeed Masiha, Saber Salehkaleybar, Niao He +2
We study the local convergence rate of stochastic first-order methods under a local -Polyak-Lojasiewicz (-PL) condition in a neighborhood of a target connected component $\ma…
Efficiently Escaping Saddle Points for Policy Optimization
Sadegh Khorasani, Saber Salehkaleybar, Negar Kiyavash +2
Policy gradient (PG) is widely used in reinforcement learning due to its scalability and good performance. In recent years, several variance-reduced PG methods have been proposed w…