arXiv subjects

Yijia Zhao

Publications and source records attributed to Yijia Zhao.

12 recordsLinked to original sources

arXiv2026

Herculean: An Agentic Benchmark for Financial Intelligence

As AI agents improve, the central question is no longer whether they can solve isolated well-defined financial tasks, but whether they can reliably carry out financial professional work. Existing financial benchmarks offer only a partial view of this ability, as they primarily evaluate static competencies such as question answering, retrieval, summarization, and classification. We introduce Herculean, the first skilled benchmark for agentic financial intelligence spanning four representative workflows, including Trading, Hedging, Market Insights, and Auditing. Each workflow is instantiated as a standardized MCP-based skill environment with its own tools, interaction dynamics, constraints, and success criteria, enabling consistent end-to-end assessment of heterogeneous agent systems. Across frontier agents, we find agents perform relatively well on Trading and Market Insights, but struggle substantially on Hedging and Auditing, where long-horizon coordination, state consistency, and structured verification are critical. Overall, our results point to a key gap in current agents in turning financial reasoning into dependable workflow execution in high-stakes financial workflows.

cs.AI↗

arXiv2026

Causal Bandit Over Unknown Graphs: Upper Confidence Bounds With Backdoor Adjustment

The causal bandit problem seeks to identify, through sequential experimentation, an intervention that maximizes the expected reward in a causal system modeled by a directed acyclic graph (DAG). Existing methods typically assume that the causal graph is known or impose restrictive structural assumptions. In this paper, we study causal bandit problems when the causal graph is unknown. We first consider Gaussian DAG models without latent confounders. By combining observational and experimental data collected sequentially during the bandit process, we identify candidate backdoor adjustment sets for each intervention arm. These sets enable estimation of causal effects and construction of upper confidence bounds that integrate information from both data sources. Based on these estimates, we propose a new algorithm, termed backdoor-adjustment upper confidence bound (BA-UCB), for sequential intervention selection. We establish finite-time upper bounds on the cumulative regret of BA-UCB, showing improved rates and substantially relaxed dependency on the number of intervention arms compared to standard multi-armed bandit methods. We further extend the methodology and theoretical guarantees to settings with latent confounders, where the observed variables are modeled by an acyclic directed mixed graph. Simulation studies demonstrate that BA-UCB achieves substantially lower cumulative regret and favorable computational efficiency relative to existing approaches.

cs.LG↗

arXiv2026

AIConfigurator: Lightning-Fast Configuration Optimization for Multi-Framework LLM Serving

Optimizing Large Language Model (LLM) inference in production systems is increasingly difficult due to dynamic workloads, stringent latency/throughput targets, and a rapidly expanding configuration space. This complexity spans not only distributed parallelism strategies (tensor/pipeline/expert) but also intricate framework-specific runtime parameters such as those concerning the enablement of CUDA graphs, available KV-cache memory fractions, and maximum token capacity, which drastically impact performance. The diversity of modern inference frameworks (e.g., TRT-LLM, vLLM, SGLang), each employing distinct kernels and execution policies, makes manual tuning both framework-specific and computationally prohibitive. We present AIConfigurator, a unified performance-modeling system that enables rapid, framework-agnostic inference configuration search without requiring GPU-based profiling. AIConfigurator combines (1) a methodology that decomposes inference into analytically modelable primitives - GEMM, attention, communication, and memory operations while capturing framework-specific scheduling dynamics; (2) a calibrated kernel-level performance database for these primitives across a wide range of hardware platforms and popular open-weights models (GPT-OSS, Qwen, DeepSeek, LLama, Mistral); and (3) an abstraction layer that automatically resolves optimal launch parameters for the target backend, seamlessly integrating into production-grade orchestration systems. Evaluation on production LLM serving workloads demonstrates that AIConfigurator identifies superior serving configurations that improve performance by up to 40% for dense models (e.g., Qwen3-32B) and 50% for MoE architectures (e.g., DeepSeek-V3), while completing searches within 30 seconds on average. Enabling the rapid exploration of vast design spaces - from cluster topology down to engine specific flags.

cs.LG↗

arXiv2025

MultiFinBen: Benchmarking Large Language Models for Multilingual and Multimodal Financial Application

Real-world financial analysis involves information across multiple languages and modalities, from reports and news to scanned filings and meeting recordings. Yet most existing evaluations of LLMs in finance remain text-only, monolingual, and largely saturated by current models. To bridge these gaps, we present MultiFinBen, the first expert-annotated multilingual (five languages) and multimodal (text, vision, audio) benchmark for evaluating LLMs in realistic financial contexts. MultiFinBen introduces two new task families: multilingual financial reasoning, which tests cross-lingual evidence integration from filings and news, and financial OCR, which extracts structured text from scanned documents containing tables and charts. Rather than aggregating all available datasets, we apply a structured, difficulty-aware selection based on advanced model performance, ensuring balanced challenge and removing redundant tasks. Evaluating 21 leading LLMs shows that even frontier multimodal models like GPT-4o achieve only 46.01% overall, stronger on vision and audio but dropping sharply in multilingual settings. These findings expose persistent limitations in multilingual, multimodal, and expert-level financial reasoning. All datasets, evaluation scripts, and leaderboards are publicly released.

cs.CL↗

arXiv2023

Flexible Coherent Optical Access: Architectures, Algorithms, and Demonstrations

To cope with the explosive bandwidth demand, significant progress has been made in the ITU-T standardization sector to define a higher-speed passive optical network (PON) with a 50Gb/s line rate. Recently, 50G PON becomes mature gradually, which means it is time to discuss beyond 50G PON. For ensuring an acceptable optical power budget, beyond 50G PON will potentially use coherent technologies, which can simultaneously promote the applications of flexible multiple access such as time/frequency-domain multiple access (TFDMA). In this paper, we will introduce the architectures, algorithms, and demonstrations for TFDMA-based coherent PON. The system architectures based on an ultra-simple coherent transceiver and specific signal spectra are designed to greatly reduce the cost of ONUs. Meanwhile, fast and low-complexity digital signal processing (DSP) algorithms are proposed for dealing with upstream and downstream signals. Based on the architectures and algorithms, we experimentally demonstrate the first real-time TFDMA-based coherent PON, which can support at most 256 end users, and peak line rates of 100Gb/s and 200Gb/s in the upstream and downstream scenarios, respectively. In conclusion, the proposed technologies for the coherent PON make it more possible to be applied in the future beyond 50G PON.

cs.NI↗

arXiv2020

Long-Distance Continuous-Variable Quantum Key Distribution over 202.81 km of Fiber

Quantum key distribution provides secure keys resistant to code-breaking quantum computers. The continuous-variable version of quantum key distribution offers the advantages of higher secret key rates in metropolitan areas, as well as the use of standard telecom components that can operate at room temperature. However, the transmission distance of these systems (compared with discrete-variable systems) are currently limited and considered unsuitable for long-distance distribution. Herein, we report the experimental results of long distance continuous-variable quantum key distribution over 202.81 km of ultralow-loss optical fiber by suitably controlling the excess noise and employing highly efficient reconciliation procedures. This record-breaking implementation of the continuous-variable quantum key distribution doubles the previous distance record and shows the road for long-distance and large-scale secure quantum key distribution using room-temperature standard telecom components.

quant-ph↗

arXiv2019

Continuous-variable QKD over 50km commercial fiber

The continuous-variable version of quantum key distribution (QKD) offers the advantages (over discrete-variable systems) of higher secret key rates in metropolitan areas as well as the use of standard telecom components that can operate at room temperature. An important step in the real-world adoption of continuous-variable QKD is the deployment of field tests over commercial fibers. Here we report two different field tests of a continuous-variable QKD system through commercial fiber networks in Xi'an and Guangzhou over distances of 30.02 km (12.48 dB) and 49.85 km (11.62 dB), respectively. We achieve secure key rates two orders-of-magnitude higher than previous field test demonstrations. This is achieved by developing a fully automatic control system to create stable excess noise and by applying a rate-adaptive reconciliation protocol to achieve a high reconciliation efficiency with high success probability. Our results pave the way to achieving continuous-variable QKD in a metropolitan setting.

quant-ph↗

arXiv2019

Polarization attack on continuous-variable quantum key distribution

The shot-noise unit (SNU) is a crucial factor for the practical security of a continuous-variable quantum key distribution system. In the most widely used experimental scheme, the SNU should be calibrated first and acts as a constant during the key distribution. However, the SNU of a practical system is dependent on the various parameters of the local oscillator, which can be controlled by the eavesdropper in the open channel. In this paper, we report a quantum hacking method to control the practical SNU by using the limited compensation rate of the polarization compensation. Since the compensation is only based on of the polarization measurement results of part of local oscillator pulses, the polarization of other unmeasured pulses may not be compensated correctly, which can be utilized by the eavesdropper to control the practical SNU. The simulation and experiment results indicate that the practical SNU can be controlled by the eavesdropper. Thus, the eavesdropper can use the fact that the practical SNU is no longer equals to the calibrated one to control the excess noise and final key rate.

quant-ph↗

arXiv2018

Continuous-variable measurement-device-independent quantum key distribution with virtual photon subtraction

The method of improving the performance of continuous-variable quantum key distribution protocols by post-selection has been recently proposed and verified. In continuous-variable measurement-device-independent quantum key distribution (CV-MDI QKD) protocols, the measurement results are obtained from untrusted third party Charlie. There is still not an effective method of improving CV-MDI QKD by the post-selection with untrusted measurement. We propose a method to improve the performance of coherent-state CV-MDI QKD protocol by non-Gaussian post-selection. The non-Gaussian post-selection of transmitted data is equivalent to an ideal photon subtraction on the two-mode squeezed vacuum state, which is favorable to enhance the performance of CV-MDI QKD. In CV-MDI QKD protocol with non-Gaussian post-selection, two users select their own data independently. We demonstrate that the optimal performance of the renovated CV-MDI QKD protocol is obtained with the transmitted data only selected by Alice. By setting appropriate parameters of the non-Gaussian post-selection, the secret key rate and tolerable excess noise are both improved at long transmission distance. The method provides an effective optimization scheme for the application of CV-MDI QKD protocols.

quant-ph↗

arXiv2017

Finite-size analysis of continuous-variable measurement-device-independent quantum key distribution

We study the impact of the finite-size effect on the continuous-variable measurement-device-independent quantum key distribution (CV-MDI QKD) protocol, mainly considering the finite-size effect on the parameter estimation procedure. The central-limit theorem and maximum likelihood estimation theorem are used to estimate the parameters. We also analyze the relationship between the number of exchanged signals and the optimal modulation variance in the protocol. It is proved that when Charlie's position is close to Bob, the CV-MDI QKD protocol has the farthest transmission distance in the finite-size scenario. Finally, we discuss the impact of finite-size effects related to the practical detection in the CV-MDI QKD protocol. The overall results indicate that the finite-size effect has a great influence on the secret key rate of the CV-MDI QKD protocol and should not be ignored.

quant-ph↗

arXiv2017

Improvement of two-way continuous-variable quantum key distribution with virtual photon subtraction

We propose a method to improve the performance of two-way continuous-variable quantum key distribution protocol by virtual photon subtraction. The Virtual photon subtraction implemented via non-Gaussian post-selection not only enhances the entanglement of two-mode squeezed vacuum state but also has advantages in simplifying physical operation and promoting efficiency. In two-way protocol, virtual photon subtraction could be applied on two sources independently. Numerical simulations show that the optimal performance of renovated two-way protocol is obtained with photon subtraction only used by Alice. The transmission distance and tolerable excess noise are improved by using the virtual photon subtraction with appropriate parameters. Moreover, the tolerable excess noise maintains a high value with the increase of distance so that the robustness of two-way continuous-variable quantum key distribution system is significantly improved, especially at long transmission distance.

quant-ph↗

arXiv2017

Numerical simulation of the optimal two-mode attacks for two-way continuous-variable quantum cryptography in reverse reconciliation

We analyze the security of the two-way continuous-variable quantum key distribution protocol in reverse reconciliation against general two-mode attacks, which represent all accessible attacks at fixed channel parameters. Rather than against one specific attack model, the expression of secret key rates of the two-way protocol are derived against all accessible attack models. It is found that there is an optimal two-mode attack to minimize the performance of the protocol in terms of both secret key rates and maximal transmission distances. We identify the optimal two-mode attack, give the specific attack model of the optimal two-mode attack and show the performance of the two-way protocol against the optimal two-mode attack. Even under the optimal two-mode attack, the performances of two-way protocol are still better than the corresponding one-way protocol, which shows the advantage of making a double use of the quantum channel and the potential of long-distance secure communication using two-way protocol.

quant-ph↗