arXiv subjects

Shaofan Yuan

Publications and source records attributed to Shaofan Yuan.

5 recordsLinked to original sources

arXiv2026

Do Implicit Personalization and Explicit Styles Conflict? PsPLUG: A Lightweight Plug-in for Balancing Personalization and Style in Customized LLMs

Personalized large language models are often expected to follow explicit style instructions, yet we find that such instructions can undermine the user-specific characteristics that personalization methods aim to preserve. We call this failure mode personalization collapse: explicit style control can conflict with implicit user preferences. To address this challenge, we propose PsPLUG, a lightweight plug-in that learns a user-specific residual after accounting for the requested style. PsPLUG also allows us to tune personalization strength at inference time. Our experiments show that explicit style instructions can diminish personalization in existing methods, whereas PsPLUG better preserves user preferences while providing precise control over the balance between personalization and style adherence.

cs.AI↗

arXiv2026

CARD: Cluster-level Adaptation with Reward-guided Decoding for Personalized Text Generation

Adapting large language models to individual users remains challenging due to the tension between fine-grained personalization and scalable deployment. We present CARD, a hierarchical framework that achieves effective personalization through progressive refinement. CARD first clusters users according to shared stylistic patterns and learns cluster-specific LoRA adapters, enabling robust generalization and strong low-resource performance. To capture individual differences within each cluster, we propose an implicit preference learning mechanism that contrasts user-authored text with cluster-level generations, allowing the model to infer user-specific style preferences without manual annotation. At inference time, CARD injects personalization exclusively at decoding via lightweight user preference vectors and low-rank logit corrections, while keeping the base model frozen. Experiments on the LaMP and LongLaMP benchmarks show that CARD achieves competitive or superior generation quality compared to state-of-the-art baselines, while significantly improving efficiency and scalability for practical personalized text generation.

cs.AI↗

arXiv2020

Moiré Band Topology in Twisted Bilayer Graphene

Recently twisted bilayer graphene (t-BLG) emerges as a new strongly correlated physical platform near a magic twist angle, which hosts many exciting phenomena such as the Mott-like insulating phases, unconventional superconducting behavior and emergent ferromagnetism. Besides the apparent significance of band flatness, band topology may be another critical element in determining strongly correlated twistronics yet receives much less attention. Here we report compelling evidence for nontrivial noninteracting band topology of t-BLG moiré Dirac bands through a systematic nonlocal transport study, in conjunction with an examination rooted in $K$-theory. The moiré band topology of t-BLG manifests itself as two pronounced nonlocal responses in the electron and hole superlattice gaps. We further show that the nonlocal responses are robust to the interlayer electric field, twist angle, and edge termination, exhibiting a universal scaling law. While an unusual symmetry of t-BLG trivializes Berry curvature, we elucidate that two $Z_2$ invariants characterize the topology of the moiré Dirac bands, validating the topological edge origin of the observed nonlocal responses. Our findings not only provide a new perspective for understanding the emerging strongly correlated phenomena in twisted van der Waals heterostructures, but also suggest a potential strategy to achieve topologically nontrivial metamaterials from topologically trivial quantum materials based on twist engineering.

cond-mat.str-el↗

arXiv2020

Strong mid-infrared photoresponse in small-twist-angle bilayer graphene

Recently the small-twist-angle (< 2°) bilayer graphene has received extraordinary attentions due to its exciting physical properties. Compared with monolayer graphene, the Brillouin zone folding in twisted bilayer graphene (TBG) leads to the formation of superlattice bandgap and significant modification of density of states. However, these emerging properties have rarely been leveraged for the realization of new optoelectronic devices. Here we demonstrate the strong, gate-tunable photoresponse in mid-infrared wavelength range of 5 to 12 um. A maximum extrinsic photoresponsivity of 26 mA/W has been achieved at 12 um when the Fermi level in a 1.81° TBG is tuned to its superlattice bandgap. Moreover, the strong photoresponse critically depends on the formation of superlattice bandgap, and it vanishes in the gapless case with ultrasmall twist angle (< 0.5°). Our demonstration reveals the promising optical properties of TBG and provides an alternative material platform for tunable mid-infrared optoelectronics.

cond-mat.mes-hall↗

arXiv2018

Efficient Electrical Detection of Mid-Infrared Graphene Plasmons at Room Temperature

Optical excitation and subsequent decay of graphene plasmons can produce a significant increase in charge-carrier temperature. An efficient method to convert this temperature elevation into a measurable electrical signal at room temperature can enable important mid-infrared applications such as thermal sensing and imaging in ubiquitous mobile devices. However, as appealing as this goal might be, it is still unrealized due to the modest thermoelectric coefficient and weak temperature-dependence of carrier transport in graphene. Here, we demonstrate mid-infrared graphene detectors consisting of arrays of plasmonic resonators interconnected by quasi one-dimensional nanoribbons. Localized barriers associated with disorder in the nanoribbons produce a dramatic temperature dependence of carrier transport, thus enabling the electrical detection of plasmon decay in the nearby graphene resonators. We further realize a device with a subwavelength footprint of 5*5 um2 operating at 12.2 um, an external responsivity of 16 mA/W, a low noise-equivalent power of 1.3 nW/Hz1/2 at room temperature, and an operational frequency potentially beyond gigahertz. Importantly, our device is fabricated using large-scale graphene and possesses a simple two-terminal geometry, representing an essential step toward the realization of on-chip graphene mid-infrared detector arrays.

cond-mat.mes-hall↗