arXiv ScienceSearch

arXiv subjects

Chenyu Fang

Publications and source records attributed to Chenyu Fang.

4 recordsLinked to original sources

Success Leaves Detours: Learning Executable Walkthroughs for Long-Horizon Agents

Test-time self-evolving agents improve by reusing past experience, yet sparse-reward trajectories contain failures, loops, and detours, while summaries often omit the state conditions and action dependencies needed for execution. We study executable Walkthrough induction from sparse-reward trajectories: extracting compact, state-conditioned, and verifiable procedures. Our key observation is that delayed credit identifies actions associated with progress but cannot determine whether they produce facts required by later actions. We propose Trace, a credit-guided, dependency-grounded framework that compiles noisy trajectories into executable Walkthrough Memory. It detects progress anchors from rewards and persistent state changes, propagates credit to identify valuable transitions, and estimates action prerequisites from cross-episode success and failure evidence. Backward dependency slicing then traces required facts to their producers, extracting dependency-consistent action chains while removing irrelevant loops and detours. The resulting Walkthroughs encode entry conditions, ordered state--action--effect steps, and completion and failure predicates, supporting reuse, intermediate-state resumption, and programmatic verification. Experiments on J-TTL, WebShop, and ScienceWorld with three open-source LLMs show that Trace consistently outperforms eight test-time learning and memory baselines. Compared with the strongest baseline, it improves average AUC and Final-$3$ by $30.0%$ and $40.5%$, respectively, while using fewer inference tokens. These results show that long-horizon interaction benefits more from state-conditioned executable procedures than from complete trajectories or abstract summaries.

cs.LG

Enhanced Charged Higgs Signal at the LHC

We investigate the discovery prospects of a charged Higgs boson ($H^\pm$) at the Large Hadron Collider (LHC) via the process $cg\to bH^\pm \to bc\bar{b}$ within the framework of a general two Higgs doublet model (G2HDM). In most two Higgs doublet models, the $H^+ cb$ coupling ($g_{H^+cb}$) is usually suppressed by the CKM matrix element $V_{cb}$. In G2HDM, there are additional Yukawa couplings, the process $cg\to bH^\pm \to bc\bar{b}$ is enhanced by the coupling $g_{H^+cb} \simeq \rtc V_{tb}$ in both the production and decay of the charged Higgs boson. We study possible physics backgrounds and evaluate the discovery potential with realistic acceptance cuts and tagging efficiencies at collider energies of $\sqrt{s}=$13 and 14 TeV. We apply $b$-tagging and $c$-tagging and show that $m_{H^+}$ can be extracted by pairing the tagged $b$ and $c$-jets. Our analysis leads to promising results for the current LHC and expected high-luminosity LHC.

hep-ph

Charming top interactions with a neutral Higgs Boson decaying into $τ^+ τ^-$ at the LHC

We investigate the discovery potential of flavor changing neutral Higgs (FCNH) interactions in top quark decays at the LHC. A general two Higgs doublet model (2HDM) is adopted to study the top quark decay $t \to c ϕ^0$, where $ϕ^0$ is either a CP-even heavy neutral scalar ($H^0$), or a CP-odd pseudoscalar ($A^0$), followed by the Higgs boson decaying into $τ^+τ^-$. The complete process is $pp \to t\bar{t} \to t cϕ^0 \to (bjj) (cτ_{lep}τ_{had}) +X$, where $t$ or $\bar{t}$ become virtual when $m_ϕ > m_t -m_c$, $τ_{lep}$ and $τ_{had}$ represent the $τ$ leptonic decay ($τ^\pm \to \ell^\pm$ + MET) and the $τ$ hadronic decay ($τ^\pm \to j_τ$ + MET), respectively. We evaluate the cross section applying realistic selection criteria for the FCNH signal and the dominant physics background at the parton-level. In addition, we employ \textsc{MadGraph}, \textsc{Pythia}~8, and \textsc{Delphes} to perform a realistic event-level Monte Carlo simulation analysis. The collinear approximation is applied since the $τ$'s from Higgs boson decays are highly boosted such that $τ_{lep}$ and $τ_{had}$ momenta are approximately in the same direction as the $τ$'s, which enables the reconstruction of the Higgs boson mass. The energy of the charm quark provides an additional kinematic variable that discriminates the signal from the background. In the alignment limit, the FCNH couplings for the heavy Higgs bosons ($H^0$ and $A^0$) will not be suppressed [$λ_{tcH} \propto \sin(β-α) \approx 1$]. We study the discovery potential for the heavy neutral Higgs bosons at the LHC with collider energies of $\sqrt{s} = 13$ TeV, 13.6 TeV and 14 TeV, and find the High-Luminosity LHC offers great promise for the discovery of heavier Higgs bosons from $t \to c ϕ^0$ or $t^* \to c ϕ^0$.

hep-ph

PlanGPT: Enhancing Urban Planning with Tailored Language Model and Efficient Retrieval

In the field of urban planning, general-purpose large language models often struggle to meet the specific needs of planners. Tasks like generating urban planning texts, retrieving related information, and evaluating planning documents pose unique challenges. To enhance the efficiency of urban professionals and overcome these obstacles, we introduce PlanGPT, the first specialized Large Language Model tailored for urban and spatial planning. Developed through collaborative efforts with institutions like the Chinese Academy of Urban Planning, PlanGPT leverages a customized local database retrieval framework, domain-specific fine-tuning of base models, and advanced tooling capabilities. Empirical tests demonstrate that PlanGPT has achieved advanced performance, delivering responses of superior quality precisely tailored to the intricacies of urban planning.

cs.CL