Source author record

Kaituo Zhang

Kaituo Zhang appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

gr-qc astro-ph.CO Artificial Intelligence Computation and Language Computer Vision hep-th Human-Computer Interaction Multimedia

Catalog footprint

What is connected

8works

8topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2026arXiv

Are Tools All We Need? Unveiling the Tool-Use Tax in LLM Agents

Tool-augmented reasoning has become a popular direction for LLM-based agents, and it is widely assumed to improve reasoning and reliability. However, we demonstrate that this consensus does not always hold: in the presence of semantic distractors, tool-augmented reasoning does not necessarily outperform native CoT. To explain this performance gap, we propose a Factorized Intervention Framework that isolates the cost of prompt formatting, the overhead of the tool-calling protocol, and the actual gain from executing tools. Our analysis reveals a critical tradeoff: under semantic noise, the gains from tools often fail to offset the "tool-use tax", which is the performance degradation introduced by the tool-calling protocol itself. To address this, we introduce G-STEP, a lightweight inference-time gate to mitigate protocol-induced errors. While this yields partial recovery, our findings suggest that more substantial improvements still require strengthening the model's intrinsic reasoning and tool-interaction capabilities.

preprint2026arXiv

Cleansing the Artificial Mind: A Self-Reflective Detoxification Framework for Large Language Models

Recent breakthroughs in Large Language Models (LLMs) have revealed remarkable generative capabilities and emerging self-regulatory mechanisms, including self-correction and self-rewarding. However, current detoxification techniques rarely exploit these built-in abilities; instead, they rely on external modules, labor-intensive data annotation, or human intervention --factors that hinder scalability and consistency. In this paper, we introduce a fully self-reflective detoxification framework that harnesses the inherent capacities of LLMs to detect, correct toxic content, and refine LLMs without external modules and data annotation. Specifically, we propose a Toxic Signal Detector --an internal self-identification mechanism, coupled with a systematic intervention process to transform toxic text into its non-toxic counterpart. This iterative procedure yields a contrastive detoxification dataset used to fine-tune the model, enhancing its ability for safe and coherent text generation. Experiments on benchmark datasets such as DetoxLLM and ParaDetox show that our method achieves better detoxification performance than state-of-the-art methods while preserving semantic fidelity. By obviating the need for human intervention or external components, this paper reveals the intrinsic self-detoxification ability of LLMs, offering a consistent and effective approach for mitigating harmful content generation. Ultimately, our findings underscore the potential for truly self-regulated language models, paving the way for more responsible and ethically guided text generation systems.

preprint2022arXiv

A Demographic Attribute Guided Approach to Age Estimation

Face-based age estimation has attracted enormous attention due to wide applications to public security surveillance, human-computer interaction, etc. With vigorous development of deep learning, age estimation based on deep neural network has become the mainstream practice. However, seeking a more suitable problem paradigm for age change characteristics, designing the corresponding loss function and designing a more effective feature extraction module still needs to be studied. What is more, change of face age is also related to demographic attributes such as ethnicity and gender, and the dynamics of different age groups is also quite different. This problem has so far not been paid enough attention to. How to use demographic attribute information to improve the performance of age estimation remains to be further explored. In light of these issues, this research makes full use of auxiliary information of face attributes and proposes a new age estimation approach with an attribute guidance module. We first design a multi-scale attention residual convolution unit (MARCU) to extract robust facial features other than simply using other standard feature modules such as VGG and ResNet. Then, after being especially treated through full connection (FC) layers, the facial demographic attributes are weight-summed by 1*1 convolutional layer and eventually merged with the age features by a global FC layer. Lastly, we propose a new error compression ranking (ECR) loss to better converge the age regression value. Experimental results on three public datasets of UTKFace, LAP2016 and Morph show that our proposed approach achieves superior performance compared to other state-of-the-art methods.

preprint2016arXiv

Running cosmological constant with observational tests

We investigate the running cosmological constant model with dark energy linearly proportional to the Hubble parameter, $Λ= σH + Λ_0$, in which the $Λ$CDM limit is recovered by taking $σ=0$. We derive the linear perturbation equations of gravity under the Friedmann-Lemaïtre-Robertson-Walker cosmology, and show the power spectra of the CMB temperature and matter density distribution. By using the Markov chain Monte Carlo method, we fit the model to the current observational data and find that $σH_0/ Λ_0 \lesssim 2.63 \times 10^{-2}$ and $6.74 \times 10^{-2}$ for $Λ(t)$ coupled to matter and radiation-matter, respectively, along with constraints on other cosmological parameters.

preprint2014arXiv

Emergent universe in spatially flat cosmological model

The scenario of an emergent universe provides a promising resolution to the big bang singularity in universes with positive or negative spatial curvature. It however remains unclear whether the scenario can be successfully implemented in a spatially flat universe which seems to be favored by present cosmological observations. In this paper, we study the stability of Einstein static state solutions in a spatially flat Shtanov-Sahni braneworld scenario. With a negative dark radiation term included and assuming a scalar field as the only matter energy component, we find that the universe can stay at an Einstein static state past eternally and then evolve to an inflation phase naturally as the scalar field climbs up its potential slowly. In addition, we also propose a concrete potential of the scalar field that realizes this scenario.

preprint2013arXiv

Oscillating universe in massive gravity

Massive gravity is a modified theory of general relativity. In this paper, we study, using a method in which the scale factor changes as a particle in a "potential", all possible cosmic evolutions in a ghost-free massive gravity. We find that there exists, in certain circumstances, an oscillating universe or a bouncing one. If the universe starts at the oscillating region, it may undergo a number of oscillations before it quantum mechanically tunnels to the bounce point and then expand forever. But going back to the singularity from the oscillating region is physically not allowed. So, the big bang singularity can be successfully resolved. At the same time, we also find that there exists a stable Einstein static state in some cases. However, the universe can not stay at this stable state past-eternally since it is allowed to quantum mechanically tunnel to a big-bang-to-big-crunch region and end with a big crunch. Thus, a stable Einstein static state universe can not be used to avoid the big bang singularity in massive gravity.

preprint2012arXiv

Oscillating universe in the DGP braneworld

With a method in which the Friedmann equation is written in a form such that evolution of the scale factor can be treated as that of a particle in a "potential", we classify all possible cosmic evolutions in the DGP braneworld scenario with the dark radiation term retained. By assuming that the energy component is pressureless matter, radiation or vacuum energy, respectively, we find that in the matter or vacuum energy dominated case, the scale factor has a minimum value $a_0$. In the matter dominated case, the big bang singularity can be avoided in some special circumstances, and there may exist an oscillating universe or a bouncing one. If the cosmic scale factor is in the oscillating region initially, the universe may undergo an oscillation. After a number of oscillations, it may evolve to the bounce point through quantum tunneling and then expand. However, if the universe contracts initially from an infinite scale, it can turn around and then expand forever. In the vacuum energy dominated case, there exists a stable Einstein static state to avoid the big bang singularity. However, in certain circumstances in the matter or vacuum energy dominated case, a new kind of singularity may occur at $a_0$ as a result of the discontinuity of the scale factor. In the radiation dominated case, the universe may originate from the big bang singularity, but a bouncing universe which avoids this singularity is also possible.

preprint2010arXiv

The stability of Einstein static universe in the DGP braneworld

The stability of an Einstein static universe in the DGP braneworld scenario is studied in this paper. Two separate branches denoted by $ε=\pm1$ of the DGP model are analyzed. Assuming the existence of a perfect fluid with a constant equation of state, $w$, in the universe, we find that, for the branch with $ε=1$, there is no a stable Einstein static solution, while, for the case with $ε=-1$, the Einstein static universe exists and it is stable when $-1<w<-1/3$. Thus, the universe can stay at this stable state past-eternally and may undergo a series of infinite, non-singular oscillations. Therefore, the big bang singularity problem in the standard cosmological model can be resolved.