arXiv subjects

Boyang Guo

Publications and source records attributed to Boyang Guo.

6 recordsLinked to original sources

arXiv2026

ECHO: A Cognitively Inspired, Auditable Memory Plane for Long-Horizon Agents

Long-horizon agents need memory that identifies relevant experience, resolves revisions, and exposes checkable provenance. We present ECHO (Embodied Context and History Orchestration), an auditable memory architecture and service prototype inspired by episodic encoding, consolidation, contextual reinstatement, reconsolidation, and executive control. This is functional inspiration, not neural equivalence; the empirical analysis focuses on retrieval and context construction. Development runs reach 96.29% Hit@10 and 73.64% turn Recall@5 on 1,536 LoCoMo category 1-4 questions, and 97.60% Hit@10, 88.84% turn Recall@5, and 88.71% session Recall@5 on all 500 LongMemEval-S questions. A five-history BEAM gate fails, and in a separate matched 91-question QA sample Mem0 OSS scores 64.84% versus ECHO's 41.76% (exact McNemar p = 0.00107), with a history-cluster interval crossing zero. A post-hoc audit found source-specific phrases in the query-expansion rules. Although no gold answer field entered the runtime, expansion-enabled retrieval scores are therefore descriptive development measurements, not independent confirmation.

cs.AI↗

arXiv2026

Cluster-Aware Neural Collapse Prompt Tuning for Long-Tailed Generalization of Vision-Language Models

Prompt learning has emerged as an efficient alternative to fine-tuning pre-trained vision-language models (VLMs). Despite its promise, current methods still struggle to maintain tail-class discriminability when adapting to class-imbalanced datasets. In this work, we propose cluster-aware neural collapse prompt tuning (CPT), which enhances the discriminability of tail classes in prompt-tuned VLMs without sacrificing their overall generalization. First, we design a cluster-invariant space by mining semantic assignments from the pre-trained VLM and mapping them to prompt-tuned features. This computes cluster-level boundaries and restricts the constraints to local neighborhoods, which reduces interference with the global semantic structure of the pre-trained VLM. Second, we introduce neural-collapse-driven discriminability optimization with three losses: textual Equiangular Tight Frame (ETF) separation loss, class-wise convergence loss, and rotation stabilization loss. These losses work together to shape intra-cluster geometry for better inter-class separation and intra-class alignment. Extensive experiments on 11 diverse datasets demonstrate that CPT outperforms SOTA methods, with stronger performance on long-tail classes and good generalization to unseen classes.

cs.CV↗

arXiv2026

ExpPortrait: Expressive Portrait Generation via Personalized Representation

While diffusion models have shown great potential in portrait generation, generating expressive, coherent, and controllable cinematic portrait videos remains a significant challenge. Existing intermediate signals for portrait generation, such as 2D landmarks and parametric models, have limited disentanglement capabilities and cannot express personalized details due to their sparse or low-rank representation. Therefore, existing methods based on these models struggle to accurately preserve subject identity and expressions, hindering the generation of highly expressive portrait videos. To overcome these limitations, we propose a high-fidelity personalized head representation that more effectively disentangles expression and identity. This representation captures both static, subject-specific global geometry and dynamic, expression-related details. Furthermore, we introduce an expression transfer module to achieve personalized transfer of head pose and expression details between different identities. We use this sophisticated and highly expressive head model as a conditional signal to train a diffusion transformer (DiT)-based generator to synthesize richly detailed portrait videos. Extensive experiments on self- and cross-reenactment tasks demonstrate that our method outperforms previous models in terms of identity preservation, expression accuracy, and temporal stability, particularly in capturing fine-grained details of complex motion.

cs.CV↗

arXiv2025

Producing type Ia supernovae from hybrid CONe white dwarfs with main-sequence binary companions at low metallicity of Z = 0.0001

The nature of progenitors of Type Ia supernovae (SNe Ia) and their explosion mechanism remain unclear. It has been suggested that SNe Ia may be resulted from thermonuclear explosions of hybrid carbon-oxygen-neon white dwarfs(CONe WDs) when they grow in mass to approach the Chandrasekhar mass limit by accreting matter from a binary main-sequence (MS) companion. In this work, we combine the results of detailed binary evolution calculations with population synthesis models to investigate the rates and delay times of SNe Ia in the CONe WD + MS channel at low metallicity environment of Z = 0.0001. For a constant star formation rate of 5 M_sun yr-1, our calculations predict that the SN Ia rates in the CONe WD + MS channel at low metallicity of Z = 0.0001 is about 0.11 - 3.89 * 10-4 yr-1. In addition,delay times in this channel cover a wide range of 0.05 - 2.5 Gyr. We further compare our results to those given by previous study for the CONe WD + MS channel with higher metallicity of Z = 0.02 to explore the influence of metallicity on the results. We find that these two metallicity environments give a slight difference in rates and delay times of SNe Ia from the CONe WD + MS channel, although SNe Ia produced at low metallicity environment of Z = 0.0001 have relatively longer delay times.

astro-ph.SR↗

arXiv2024

The Red Supergiant Progenitor of the Type II Supernova 2024abfl

Linkage between core-collapse supernovae (SNe) and their progenitors is not fully understood and ongoing effort of searching and identifying the progenitors is needed. $\mathrm{SN\,2024abfl}$ is a recent Type II supernova exploded in the nearby star-bursting galaxy $\mathrm{NGC\,2146}$, which is also the host galaxy of $\mathrm{SN\,2018zd}$. From archival Hubble Space Telescope (HST) data, we have found a red source ($\mathrm{m_{F814W} \sim 25}$) near the location (angular distance $\leq 0.2"$) of $\mathrm{SN\,2024abfl}$ before its explosion. With F814W and F606W photometry, we found that the properties of this source matched a typical red supergiant (RSG) moderately reddened by interstellar dust at the distance of the host galaxy. We conclude that the $\mathrm{SN\,2024abfl}$ had an RSG progenitor with initial mass of $\mathrm{10M_{\odot}}$--$\mathrm{16\,M_{\odot}}$.

astro-ph.SR↗

arXiv2024

One Shot, One Talk: Whole-body Talking Avatar from a Single Image

Building realistic and animatable avatars still requires minutes of multi-view or monocular self-rotating videos, and most methods lack precise control over gestures and expressions. To push this boundary, we address the challenge of constructing a whole-body talking avatar from a single image. We propose a novel pipeline that tackles two critical issues: 1) complex dynamic modeling and 2) generalization to novel gestures and expressions. To achieve seamless generalization, we leverage recent pose-guided image-to-video diffusion models to generate imperfect video frames as pseudo-labels. To overcome the dynamic modeling challenge posed by inconsistent and noisy pseudo-videos, we introduce a tightly coupled 3DGS-mesh hybrid avatar representation and apply several key regularizations to mitigate inconsistencies caused by imperfect labels. Extensive experiments on diverse subjects demonstrate that our method enables the creation of a photorealistic, precisely animatable, and expressive whole-body talking avatar from just a single image.

cs.CV↗