Kedi Zhang | Searcharxiv

arXiv2026

Constraining the Potential Index $n$ of the Early Dark Energy Model Using Cosmic Birefringence from Planck and ACT

Cosmic birefringence and the Hubble tension represent compelling challenges to the standard $\Lambda$CDM model. The early dark energy (EDE) model with potentials $V(\phi) \propto [1-\cos(\phi/f)]^n$ offer a unified framework to address both anomalies through energy injection near matter-radiation equality and parity-violating Chern--Simons coupling to photons. While previous studies have focused on $n=3$, the dependence of the birefringence signal on the potential index $n$ remains largely unexplored. We perform a comprehensive statistical analysis of axion-like EDE models with $n=2$, $n=3$, and $n=\infty$, using $EB$ cross-polarization data from Planck-$EB$ and ACT-$EB$. The $n=2$ model is severely disadvantaged, displaying extreme coupling values ($gM_{\rm pl} \approx 69.912$ for Planck, $-40.726$ for ACT), large $\chi^2_{\rm min}$ (144.52 and 86.93), and $\Delta \chi^2<1$ with many local minials. Conversely, $n=3$ achieves the best fits ($\chi^2_{\rm min} = 65.70$ and $48.08$) with consistent couplings ($gM_{\rm pl} = -0.210 \pm 0.024$ and $-0.158 \pm 0.025$) that accurately reproduce observations across all angular scales. We checked that the $n=3$ configuration represents the optimal choice for simultaneously addressing the Hubble tension and cosmic birefringence within a theoretically viable framework.

astro-ph.CO↗

arXiv2025

Dynamic Embedding of Hierarchical Visual Features for Efficient Vision-Language Fine-Tuning

Large Vision-Language Models (LVLMs) commonly follow a paradigm that projects visual features and then concatenates them with text tokens to form a unified sequence input for Large Language Models (LLMs). However, this paradigm leads to a significant increase in the length of the input sequence, resulting in substantial computational overhead. Existing methods attempt to fuse visual information into the intermediate layers of LLMs, which alleviate the sequence length issue but often neglect the hierarchical semantic representations within the model and the fine-grained visual information available in the shallower visual encoding layers. To address this limitation, we propose DEHVF, an efficient vision-language fine-tuning method based on dynamic embedding and fusion of hierarchical visual features. Its core lies in leveraging the inherent hierarchical representation characteristics of visual encoders and language models. Through a lightweight hierarchical visual fuser, it dynamically selects and fuses hierarchical features corresponding to semantic granularity based on the internal representations of each layer in LLMs. The fused layer-related visual features are then projected and aligned before being directly embedded into the Feed-Forward Network (FFN) of the corresponding layer in LLMs. This approach not only avoids sequence expansion but also dynamically fuses multi-layer visual information. By fine-tuning only a small number of parameters, DEHVF achieves precise alignment and complementarity of cross-modal information at the same semantic granularity. We conducted experiments across various VL benchmarks, including visual question answering on ScienceQA and image captioning on COCO Captions. The results demonstrate that DEHVF achieves higher accuracy than existing parameter-efficient fine-tuning (PEFT) baselines while maintaining efficient training and inference.

cs.CV↗