Searcharxiv⌕ Search

arXiv subjects

Jianbin Chen

Publications and source records attributed to Jianbin Chen.

2 recordsLinked to original sources

EbbFlow: PINN-Inspired Progressive Curriculum for MeanFlow Training

MeanFlow learns interval-average velocity for few-step generation, but its fixed temporal sampler mixes diagonal and interior supervision throughout training. We interpret temporal coordinates as collocation points: diagonal samples provide a stochastic boundary target, whereas finite intervals supply a differential-consistency target. The diagonal fraction controls how often the consistency component appears, while interval length enters its gradient through an explicit factor. Motivated by this distinction, EbbFlow uses Temporal Supervision Allocation (TSA) to schedule the diagonal fraction and interval geometry. It starts with diagonal supervision, then admits interior samples and expands their maximum interval; a left-anchored interval law preserves coverage across time. The per-sample loss, target, backbone, and inference procedure remain unchanged. In matched class-conditional ImageNet $256\times 256$ runs without classifier-free guidance (DiT-B/2, one training seed per method, 400k updates), EbbFlow achieves 50k-sample FID of 39.65 at one network evaluation and 36.02 at two-evaluation ODE sampling, versus 44.82 and 39.86 for MeanFlow. Gradient probes show a mid-training difference in component geometry. Ablations support allocation scheduling and left-anchored intervals, although a fixed full-horizon variant performs better at one evaluation. These results suggest that temporal sampling is a useful training control whose benefits depend on the inference protocol.

cs.LG↗

Design and analysis for constrained order-of-addition experiments

In an order-of-addition (OofA) experiment, the sequence of m different components can significantly impact the experiment's response. In many OofA experiments, the components are subject to constraints, where certain orders are impossible. For example, in survey design and job scheduling, the components are often arranged into groups, and these groups of components must be placed in a fixed order. If two components are in different groups, their pairwise order is determined by the fixed order of their groups. Design and analysis are needed for these pairwise-group constrained OofA experiments. A new model is proposed to accommodate pairwise-group constraints. This paper also introduces a model for mixed-pairwise constrained OofA experiments, which allows one pair of components within each group to have a pre-determined pairwise order. It is proven that the full design, which uses all feasible orders exactly once, is D- and G-optimal under the proposed models. Systematic construction methods are used to find optimal fractional designs for pairwise-group and mixed-pairwise constrained OofA experiments. The proposed methods efficiently assess the impact of question order in a survey dataset, where participants answered generalized intelligence questions in a randomly assigned order under mixed-pairwise constraints.

stat.ME↗