arXiv subjects

Lunjun Liu

Publications and source records attributed to Lunjun Liu.

4 recordsLinked to original sources

arXiv2026

Can Large Language Models Simulate Human Cognition Beyond Behavioral Imitation?

An essential problem in artificial intelligence is whether LLMs can simulate human cognition or merely imitate surface-level behaviors, while existing datasets suffer from either synthetic reasoning traces or population-level aggregation, failing to capture authentic individual cognitive patterns. We introduce a benchmark grounded in the longitudinal research trajectories of 217 researchers across diverse domains of artificial intelligence, where each author's scientific publications serve as an externalized representation of their cognitive processes. To distinguish whether LLMs transfer cognitive patterns or merely imitate behaviors, our benchmark deliberately employs a cross-domain, temporal-shift generalization setting. A multidimensional cognitive alignment metric is further proposed to assess individual-level cognitive consistency. Through systematic evaluation of state-of-the-art LLMs and various enhancement techniques, we provide a first-stage empirical study on the questions: (1) How well do current LLMs simulate human cognition? and (2) How far can existing techniques enhance these capabilities?

cs.CL↗

arXiv2025

RSMLP: A light Sampled MLP Structure for Incomplete Utterance Rewrite

The Incomplete Utterance Rewriting (IUR) task has garnered significant attention in recent years. Its goal is to reconstruct conversational utterances to better align with the current context, thereby enhancing comprehension. In this paper, we introduce a novel and versatile lightweight method, Rewritten-Sampled MLP (RSMLP). By employing an MLP based architecture with a carefully designed down-sampling strategy, RSMLP effectively extracts latent semantic information between utterances and makes appropriate edits to restore incomplete utterances. Due to its simple yet efficient structure, our method achieves competitive performance on public IUR datasets and in real-world applications.

cs.CL↗

arXiv2024

Tacit Learning with Adaptive Information Selection for Cooperative Multi-Agent Reinforcement Learning

In multi-agent reinforcement learning (MARL), the centralized training with decentralized execution (CTDE) framework has gained widespread adoption due to its strong performance. However, the further development of CTDE faces two key challenges. First, agents struggle to autonomously assess the relevance of input information for cooperative tasks, impairing their decision-making abilities. Second, in communication-limited scenarios with partial observability, agents are unable to access global information, restricting their ability to collaborate effectively from a global perspective. To address these challenges, we introduce a novel cooperative MARL framework based on information selection and tacit learning. In this framework, agents gradually develop implicit coordination during training, enabling them to infer the cooperative behavior of others in a discrete space without communication, relying solely on local information. Moreover, we integrate gating and selection mechanisms, allowing agents to adaptively filter information based on environmental changes, thereby enhancing their decision-making capabilities. Experiments on popular MARL benchmarks show that our framework can be seamlessly integrated with state-of-the-art algorithms, leading to significant performance improvements.

cs.MA↗

arXiv2021

Probing Intra-Halo Light with Galaxy Stacking in CIBER Images

We study the stellar halos of $0.2\lesssim z \lesssim 0.5$ galaxies with stellar masses spanning $M_*\sim 10^{10.5}$ to $10^{12}M_\odot$ (approximately $L_*$ galaxies at this redshift) using imaging data from the Cosmic Infrared Background Experiment (CIBER). A previous CIBER fluctuation analysis suggested that intra-halo light (IHL) contributes a significant portion of the near-infrared extragalactic background light (EBL), the integrated emission from all sources throughout cosmic history. In this work, we carry out a stacking analysis with a sample of $\sim$30,000 Sloan Digital Sky Survey (SDSS) photometric galaxies from CIBER images in two near-infrared bands (1.1 and 1.8 $\mu$m) to directly probe the IHL associated with these galaxies. We stack galaxies in five sub-samples split by brightness, and detect an extended galaxy profile, beyond the instrument point spread function (PSF), derived by stacking stars. We jointly fit a model for the inherent galaxy light profile, plus large-scale one- and two-halo clustering to measure the extended galaxy IHL. We detect non-linear one-halo clustering in the 1.8 $\mu$m band, at a level consistent with numerical simulations. Our results on the galaxy profile suggest that $\sim 50\%$ of the total galaxy light budget in our galaxy sample resides in the outskirts of the galaxies at $r > 10$ kpc. We describe this extended emission as IHL and and are able to study how this fraction evolves with cosmic time. These results are new in the near-infrared wavelength at the $L_*$ mass scale, and suggest that IHL has a significant contribution to the integrated galactic light, and to the amplitude of large-scale background fluctuations.

astro-ph.GA↗