arXiv subjects

Bo Qu

Publications and source records attributed to Bo Qu.

17 recordsLinked to original sources

arXiv2026

The Horizon Gap: Planning, Memory, Execution, Training, and Evaluation for Long-Horizon LLM Agents

Frontier language models solve reasoning problems in a single forward pass that would have been research contributions years ago, yet fail at multi-hour tasks: losing track of earlier decisions, declaring half-finished work done, or drifting from goals. We call this the horizon gap and survey 1,547 arXiv papers (2024-2026) collected via systematic seed harvest with a disclosed 26.8% bleed filter, extended by targeted supplementation. We disambiguate three routinely conflated properties: long-horizon (task property: required steps), long-context (model property: token capacity), and long-term memory (system property: persistence across steps/sessions). We organize the corpus into six categories tracking a long-horizon task's lifecycle -- planning, memory, execution, training, evaluation, and foundations/safety -- crossed with an axis capturing where horizons are carried (within-context, within-task-beyond-context, or cross-task-persistent). Across all categories, we find the same pattern: outcome-only signals grow uninformative as horizons lengthen, and the field's response -- whether process reward models, credit assignment, or trajectory-level diagnostics -- manufactures denser step-level signals. We treat critical and diagnostic literature as first-class threads throughout, arguing that segregating critique from method would routinely split single papers across chapters. We close by naming open measurement problems: decomposing model versus harness capability, managing correlated bias in process-level signals used for both training and evaluation, and whether long-horizon reliability admits general predictive theory.

cs.CL↗

arXiv2026

The Calibration Floor: Format Repair Can Masquerade as Self-Correction at Small-to-Mid Scale

Accuracy changes after language-model self-revision are usually interpreted as changes in reasoning. We show this can fail at the answer-extraction boundary, and test the failure causally rather than only observationally. Across Qwen3.5 (0.8B-9B), Gemma-4-12B, and two frontier models via API (Tencent Hy3, Nvidia Nemotron-3-Ultra-550B) in 29 primary cells plus a frontier arm, we decompose the always-revise accuracy shift into a content margin (both answers parseable) and format-recovery/loss margins (parseability changes). On 12 cells with meaningful unparseable-answer rates, format effects exceed content effects (Wilcoxon p=1.7e-3). To test this causally, we force already-generated reasoning through grammar-constrained decoding so every answer is parseable by construction: across 14 cells this closes a median 71% of the gap between the naive total effect and the content-margin estimate, with two cells converging exactly and a residual on the two largest-effect cells reported rather than dismissed. A clustered model confirms floor-scale (0.8B/2B) models have far higher odds of content-level change and harm than capable-scale models (p<1e-7). Replicating a cited confidence-gating protocol verbatim on Qwen3.5 does not reproduce its reported gain and shows the same near-zero content margin. A frontier check on much larger models shows format-dominance intensifying with scale: content margin is exactly zero in all 5 cells despite total effects up to +0.275, though this arm is lower-powered. The calibration-floor criterion on the content margin reveals a squeeze: floor-scale cells have headroom but insufficient signal, capable-scale cells have signal but little headroom; only one cell is marginally viable, with negligible sealed-holdout gain. Content is a minority share of what the field has measured as self-correction. We release the instrument, code, and derived results.

cs.CL↗

arXiv2026

Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loops

AI systems increasingly participate in their own improvement: revising their outputs, adapting their harnesses during deployment, training on data they generate, and conducting AI research itself. This literature uses a vocabulary ("self-refine," "self-reward," "self-play," "self-evolve") that conflates fundamentally different ambitions. We survey 1,250 arXiv papers (2024-2026) along two axes: what the system improves -- its behavior in deployment, its policy through training, its evaluator, or the research process itself -- and the degree of loop closure (human-in-the-loop to fully closed). The taxonomy separates bounded self-refinement -- convergent, evaluable, and industrial practice -- from open-ended recursive self-improvement (RSI), which remains bounded by grounding requirements, collapse dynamics, and compute constraints on every measured axis. Its distinctive feature is a dedicated category for self-evaluation: every improvement loop is a claim that some signal can substitute for human judgment. We survey the evaluator design space -- judges, process reward models, verifiers, rubrics, meta-evaluation -- order the signals into a verification hierarchy from formal verifiers (strongest) to intrinsic self-assessment (weakest), and observe that demonstrated self-improvement strength tracks this hierarchy, that its failure modes (self-confirming loops, model and diversity collapse) follow from its violations, and that the "research direction-setting" bottleneck keeping humans in the loop divides into a verification problem the hierarchy indexes and a prior one -- choosing what deserves evaluation at all -- that it does not. We connect the literature to the theory of RSI limits and to the safety and governance questions raised by frontier-lab accounts of closing the loop, and identify governance-grade measurement of self-improvement as the field's most underpopulated niche.

cs.AI↗

arXiv2026

CLQT: A Closed-Loop, Cost-Aware, Strategy-Consistent Benchmark for Diagnostic Evaluation of LLM Portfolio-Management Agents

LLM agents are increasingly cast as autonomous portfolio managers, and benchmarks have moved from financial question-answering to sequential trading. Yet most still rank agents by returns over a fixed window, a weak proxy: the market path dominates a period's return, and apparent alpha can dissolve once look-ahead leakage is controlled. We introduce CLQT, which reframes closed-loop trading evaluation as diagnosis before ranking: an instrument that localizes where and why an agent's process succeeds or fails. CLQT is a closed-loop, cost-aware, strategy-consistent, temporally-gated environment whose agents run a five-stage cycle: gather, synthesize, allocate, execute, reflect. Every round emits a DecisionRound sealed into a recompute-verifiable hash chain. Six pillars form the substrate: a TimeGate, transaction- and financing-cost modeling, strategy-consistency scoring, three-tier memory, a Model-Context-Protocol tool layer, and mandate-aware synthesis. The same agent runs structured or fully autonomous, making process scaffolding an experimental variable. From the audit trail we compute a five-axis capability scorecard (APM-CS: Coherence, Acuity, Composure, Discipline, Reliability), with Coherence partly judged by a held-out, out-of-cohort LLM. We validate CLQT on a contamination-controlled multi-model backtest with an ablation grid and a live broker track on unseen, post-cutoff data, against a repeated-run noise floor. Three findings illustrate the diagnosis-first read: the capability leader is not the Sharpe leader (the nominal winner is a 5-of-26-round reliability artifact); signal-action agreement exceeds a held-out judge's coherence score by a stable margin on both tracks (+0.30 backtest, +0.23 live); and module value registers on the capability axes when returns cannot separate it. CLQT yields not a model ranking but a durable, extensible map of agent competencies and limitations.

cs.AI↗

arXiv2026

InvestPhilBench: A Multi-Layer Benchmark for Evaluating Large Language Model Procedural Reasoning in Expert Investment Philosophy

Large language models are increasingly deployed as investment research assistants, yet no benchmark tests whether they can accurately reconstruct and apply the specific procedural decision frameworks of expert investors. We introduce InvestPhilBench, a multi-layer benchmark spanning eight cognitive tiers, from principle identification (L1) to novel framework extrapolation (L8). The v0.6 release comprises 118 primary-source-verified principle cards, 25 decision-framework cards with explicit topology metadata, and 243 QA questions (197 dev / 46 held-out test). For reproducible scoring at scale we introduce the Benchmark Automated Scoring Pipeline (BASP), five algorithmic metrics, the Failure Mode Detection Protocol (FMDP) covering six failure modes, and Gate Reconstruction Accuracy (GRA), a per-gate metric for questions with gold reasoning programs. This release is primarily a benchmark-and-methodology contribution: its empirical study -- a four-model sanity wave on the 188-question development split (closed-book) -- is deliberately preliminary and stress-tests the metric design rather than ranking models. The wave shows a sharp provider-tier split (BASP 0.906 vs. 0.438), though these mixed-judge numbers are confounded upper bounds. The central methodological finding survives the caveat: the BASP composite saturates at the frontier (Claude L4 = 0.932) while GRA still exposes a procedural deficit (frontier L4 GRA ~0.77, L7 GRA 0.57-0.62) -- composite scoring rewards fluent prose and hides the procedural gap. On a 100-item expert-annotated gold set, the automated BASP composite tracks the human reference at Pearson r = 0.72 (MAE = 0.10). v0.6 also implements a unified judge and true model-in-the-loop retrieval/oracle conditions; the de-confounded multi-model leaderboard and full three-condition run are v1.0 deliverables.

cs.AI↗

arXiv2025

LLM-Enhanced Self-Evolving Reinforcement Learning for Multi-Step E-Commerce Payment Fraud Risk Detection

This paper presents a novel approach to e-commerce payment fraud detection by integrating reinforcement learning (RL) with Large Language Models (LLMs). By framing transaction risk as a multi-step Markov Decision Process (MDP), RL optimizes risk detection across multiple payment stages. Crafting effective reward functions, essential for RL model success, typically requires significant human expertise due to the complexity and variability in design. LLMs, with their advanced reasoning and coding capabilities, are well-suited to refine these functions, offering improvements over traditional methods. Our approach leverages LLMs to iteratively enhance reward functions, achieving better fraud detection accuracy and demonstrating zero-shot capability. Experiments with real-world data confirm the effectiveness, robustness, and resilience of our LLM-enhanced RL framework through long-term evaluations, underscoring the potential of LLMs in advancing industrial RL applications.

cs.LG↗

arXiv2025

EDFFDNet: Towards Accurate and Efficient Unsupervised Multi-Grid Image Registration

Previous deep image registration methods that employ single homography, multi-grid homography, or thin-plate spline often struggle with real scenes containing depth disparities due to their inherent limitations. To address this, we propose an Exponential-Decay Free-Form Deformation Network (EDFFDNet), which employs free-form deformation with an exponential-decay basis function. This design achieves higher efficiency and performs well in scenes with depth disparities, benefiting from its inherent locality. We also introduce an Adaptive Sparse Motion Aggregator (ASMA), which replaces the MLP motion aggregator used in previous methods. By transforming dense interactions into sparse ones, ASMA reduces parameters and improves accuracy. Additionally, we propose a progressive correlation refinement strategy that leverages global-local correlation patterns for coarse-to-fine motion estimation, further enhancing efficiency and accuracy. Experiments demonstrate that EDFFDNet reduces parameters, memory, and total runtime by 70.5%, 32.6%, and 33.7%, respectively, while achieving a 0.5 dB PSNR gain over the state-of-the-art method. With an additional local refinement stage,EDFFDNet-2 further improves PSNR by 1.06 dB while maintaining lower computational costs. Our method also demonstrates strong generalization ability across datasets, outperforming previous deep learning methods.

cs.CV↗

arXiv2025

ArtiMuse: Fine-Grained Image Aesthetics Assessment with Joint Scoring and Expert-Level Understanding

The rapid advancement of educational applications, artistic creation, and AI-generated content (AIGC) technologies has substantially increased practical requirements for comprehensive Image Aesthetics Assessment (IAA), particularly demanding methods capable of delivering both quantitative scoring and professional understanding. Multimodal Large Language Model (MLLM)-based IAA methods demonstrate stronger perceptual and generalization capabilities compared to traditional approaches, yet they suffer from modality bias (score-only or text-only) and lack fine-grained attribute decomposition, thereby failing to support further aesthetic assessment. In this paper, we present:(1) ArtiMuse, an innovative MLLM-based IAA model with Joint Scoring and Expert-Level Understanding capabilities; (2) ArtiMuse-10K, the first expert-curated image aesthetic dataset comprising 10,000 images spanning 5 main categories and 15 subcategories, each annotated by professional experts with 8-dimensional attributes analysis and a holistic score. Both the model and dataset will be made public to advance the field.

cs.CV↗

arXiv2024

Multi-task CNN Behavioral Embedding Model For Transaction Fraud Detection

The burgeoning e-Commerce sector requires advanced solutions for the detection of transaction fraud. With an increasing risk of financial information theft and account takeovers, deep learning methods have become integral to the embedding of behavior sequence data in fraud detection. However, these methods often struggle to balance modeling capabilities and efficiency and incorporate domain knowledge. To address these issues, we introduce the multitask CNN behavioral Embedding Model for Transaction Fraud Detection. Our contributions include 1) introducing a single-layer CNN design featuring multirange kernels which outperform LSTM and Transformer models in terms of scalability and domain-focused inductive bias, and 2) the integration of positional encoding with CNN to introduce sequence-order signals enhancing overall performance, and 3) implementing multitask learning with randomly assigned label weights, thus removing the need for manual tuning. Testing on real-world data reveals our model's enhanced performance of downstream transaction models and comparable competitiveness with the Transformer Time Series (TST) model.

cs.LG↗

arXiv2021

MuVAM: A Multi-View Attention-based Model for Medical Visual Question Answering

Medical Visual Question Answering (VQA) is a multi-modal challenging task widely considered by research communities of the computer vision and natural language processing. Since most current medical VQA models focus on visual content, ignoring the importance of text, this paper proposes a multi-view attention-based model(MuVAM) for medical visual question answering which integrates the high-level semantics of medical images on the basis of text description. Firstly, different methods are utilized to extract the features of the image and the question for the two modalities of vision and text. Secondly, this paper proposes a multi-view attention mechanism that include Image-to-Question (I2Q) attention and Word-to-Text (W2T) attention. Multi-view attention can correlate the question with image and word in order to better analyze the question and get an accurate answer. Thirdly, a composite loss is presented to predict the answer accurately after multi-modal feature fusion and improve the similarity between visual and textual cross-modal features. It consists of classification loss and image-question complementary (IQC) loss. Finally, for data errors and missing labels in the VQA-RAD dataset, we collaborate with medical experts to correct and complete this dataset and then construct an enhanced dataset, VQA-RADPh. The experiments on these two datasets show that the effectiveness of MuVAM surpasses the state-of-the-art method.

cs.CV↗

arXiv2021

Deep Attributed Network Representation Learning via Attribute Enhanced Neighborhood

Attributed network representation learning aims at learning node embeddings by integrating network structure and attribute information. It is a challenge to fully capture the microscopic structure and the attribute semantics simultaneously, where the microscopic structure includes the one-step, two-step and multi-step relations, indicating the first-order, second-order and high-order proximity of nodes, respectively. In this paper, we propose a deep attributed network representation learning via attribute enhanced neighborhood (DANRL-ANE) model to improve the robustness and effectiveness of node representations. The DANRL-ANE model adopts the idea of the autoencoder, and expands the decoder component to three branches to capture different order proximity. We linearly combine the adjacency matrix with the attribute similarity matrix as the input of our model, where the attribute similarity matrix is calculated by the cosine similarity between the attributes based on the social homophily. In this way, we preserve the second-order proximity to enhance the robustness of DANRL-ANE model on sparse networks, and deal with the topological and attribute information simultaneously. Moreover, the sigmoid cross-entropy loss function is extended to capture the neighborhood character, so that the first-order proximity is better preserved. We compare our model with the state-of-the-art models on five real-world datasets and two network analysis tasks, i.e., link prediction and node classification. The DANRL-ANE model performs well on various networks, even on sparse networks or networks with isolated nodes given the attribute information is sufficient.

cs.AI↗

arXiv2017

Modeling of Information Diffusion on Social Networks with Applications to WeChat

Traces of user activities recorded in online social networks such as the creation, viewing and forwarding/sharing of information over time open new possibilities to quantitatively and systematically understand the information diffusion process on social networks. From an online social network like WeChat, we could collect a large number of information cascade trees, each of which tells the spreading trajectory of a message/information such as which user creates the information and which users view or forward the information shared by which neighbors. In this work, we propose two heterogeneous non-linear models. Both models are validated by the WeChat data in reproducing and explaining key features of cascade trees. Specifically, we firstly apply the Random Recursive Tree (RRT) to model the cascade tree topologies, capturing key features, i.e. the average path length and degree variance of a cascade tree in relation to the number of nodes (size) of the tree. The RRT model with a single parameter $\theta$ describes the growth mechanism of a tree, where a node in the existing tree has a probability $d_i^{\theta}$ of being connected to a newly added node that depends on the degree $d_i$ of the existing node. The identified parameter $\theta$ quantifies the relative depth or broadness of the cascade trees, indicating that information propagates via a star-like broadcasting or viral-like hop by hop spreading. The RRT model explains the appearance of hubs, thus a possibly smaller average path length as the cascade size increases, as observed in WeChat. We further propose the stochastic Susceptible View Forward Removed (SVFR) model to depict the dynamic user behaviors including creating, viewing, forwarding and ignoring a message on a given social network.

cs.SI↗

arXiv2016

The Accuracy of Mean-Field Approximation for Susceptible-Infected-Susceptible Epidemic Spreading

The epidemic spreading has been studied for years by applying the mean-field approach in both homogeneous case, where each node may get infected by an infected neighbor with the same rate, and heterogeneous case, where the infection rates between different pairs of nodes are different. Researchers have discussed whether the mean-field approaches could accurately describe the epidemic spreading for the homogeneous cases but not for the heterogeneous cases. In this paper, we explore under what conditions the mean-field approach could perform well when the infection rates are heterogeneous. In particular, we employ the Susceptible-Infected-Susceptible (SIS) model and compare the average fraction of infected nodes in the metastable state obtained by the continuous-time simulation and the mean-field approximation. We concentrate on an individual-based mean-field approximation called the N-intertwined Mean Field Approximation (NIMFA), which is an advanced approach considered the underlying network topology. Moreover, we consider not only the independent and identically distributed (i.i.d.) infection rate but also the infection rate correlated with the degree of the two end nodes. We conclude that NIMFA is generally more accurate when the prevalence of the epidemic is higher. Given the same effective infection rate, NIMFA is less accurate when the variance of the i.i.d.\ infection rate or the correlation between the infection rate and the nodal degree leads to a lower prevalence. Moreover, given the same actual prevalence, NIMFA performs better in the cases: 1) when the variance of the i.i.d.\ infection rates is smaller (while the average is unchanged); 2) when the correlation between the infection rate and the nodal degree is positive.

physics.soc-ph↗

arXiv2016

SIS Epidemic Spreading with Correlated Heterogeneous Infection Rates

The epidemic spreading has been widely studied when each node may get infected by an infected neighbor with the same rate. However, the infection rate between a pair of nodes is usually heterogeneous and even correlated with their nodal degrees in the contact network. We aim to understand how such correlated heterogeneous infection rates influence the spreading on different network topologies. Motivated by real-world datasets, we propose a correlated heterogeneous Susceptible-Infected-Susceptible model which assumes that the infection rate $\beta_{ij}(=\beta_{ji})$ between node $i$ and $j$ is correlated with the degree of the two end nodes: $\beta_{ij}=c(d_id_j)^\alpha$, where $\alpha$ indicates the strength of the correlation and $c$ is selected so that the average infection rate is $1$. In order to understand the effect of such correlation on epidemic spreading, we consider as well the corresponding uncorrected but still heterogeneous infection rate scenario as a reference, where the original correlated infection rates in our CSIS model are shuffled and reallocated to the links of the same network topology. We compare these two scenarios in the average fraction of infected nodes in the metastable state on Erd{\"o}s-R{\'e}nyi (ER) and scale-free (SF) networks with a similar average degree. Through the continuous-time simulations, we find that, when the recovery rate is small, the negative correlation is more likely to help the epidemic spread and the positive correlation prohibit the spreading; as the recovery rate increases to be larger than a critical value, the positive but not negative correlation tends to help the spreading. Our findings are further analytically proved in a wheel network (one central node connects with each of the nodes in a ring) and validated on real-world networks with correlated heterogeneous interaction frequencies.

physics.soc-ph↗

arXiv2015

SIS Epidemic Spreading with Heterogeneous Infection Rates

In this work, we aim to understand the influence of the heterogeneity of infection rates on the Susceptible-Infected-Susceptible (SIS) epidemic spreading. Employing the classic SIS model as the benchmark, we study the influence of the independently identically distributed infection rates on the average fraction of infected nodes in the metastable state. The log-normal, gamma and a newly designed distributions are considered for infection rates. We find that, when the recovery rate is small, i.e.\ the epidemic spreads out in both homogeneous and heterogeneous cases: 1) the heterogeneity of infection rates on average retards the virus spreading, and 2) a larger even-order moment of the infection rates leads to a smaller average fraction of infected nodes, but the odd-order moments contribute in the opposite way; when the recovery rate is large, i.e.\ the epidemic may die out or infect a small fraction of the population, the heterogeneity of infection rates may enhance the probability that the epidemic spreads out. Finally, we verify our conclusions via real-world networks with their heterogeneous infection rates. Our results suggest that, in reality the epidemic spread may not be so severe as the classic SIS model indicates, but to eliminate the epidemic is probably more difficult.

physics.soc-ph↗

arXiv2014

Heterogeneous Recovery Rates against SIS Epidemics in Directed Networks

The nodes in communication networks are possibly and most likely equipped with different recovery resources, which allow them to recover from a virus with different rates. In this paper, we aim to understand know how to allocate the limited recovery resources to efficiently prevent the spreading of epidemics. We study the susceptible-infected-susceptible (SIS) epidemic model on directed scale-free networks. In the classic SIS model, a susceptible node can be infected by an infected neighbor with the infection rate $β$ and an infected node can be recovered to be susceptible again with the recovery rate $δ$. In the steady state a fraction $y_\infty$ of nodes are infected, which shows how severely the network is infected. We propose to allocate the recovery rate $δ_i$ for node $i$ according to its indegree and outdegree-$δ_i\scriptsize{\sim}k_{i,in}^{α_{in}}k_{i,out}^{α_{out}}$, given the finite average recovery rate $\langleδ\rangle$ representing the limited recovery resources over the whole network. We find that, by tuning the two scaling exponents $α_{in}$ and $α_{out}$, we can always reduce the infection fraction $y_\infty$ thus reducing the extent of infections, comparing to the homogeneous recovery rates allocation. Moreover, we can find our optimal strategy via the optimal choice of the exponent $α_{in}$ and $α_{out}$. Our optimal strategy indicates that when the recovery resources are sufficient, more resources should be allocated to the nodes with a larger indegree or outdegree, but when the recovery resource is very limited, only the nodes with a larger outdegree should be equipped with more resources. We also find that our optimal strategy works better when the recovery resources are sufficient but not yet able to make the epidemic die out, and when the indegree outdegree correlation is small.

physics.soc-ph↗

arXiv2014

Non-consensus opinion model on directed networks

Dynamic social opinion models have been widely studied on undirected networks, and most of them are based on spin interaction models that produce a consensus. In reality, however, many networks such as Twitter and the World Wide Web are directed and are composed of both unidirectional and bidirectional links. Moreover, from choosing a coffee brand to deciding who to vote for in an election, two or more competing opinions often coexist. In response to this ubiquity of directed networks and the coexistence of two or more opinions in decision-making situations, we study a non-consensus opinion model introduced by Shao et al. \cite{shao2009dynamic} on directed networks. We define directionality $\xi$ as the percentage of unidirectional links in a network, and we use the linear correlation coefficient $\rho$ between the indegree and outdegree of a node to quantify the relation between the indegree and outdegree. We introduce two degree-preserving rewiring approaches which allow us to construct directed networks that can have a broad range of possible combinations of directionality $\xi$ and linear correlation coefficient $\rho$ and to study how $\xi$ and $\rho$ impact opinion competitions. We find that, as the directionality $\xi$ or the indegree and outdegree correlation $\rho$ increases, the majority opinion becomes more dominant and the minority opinion's ability to survive is lowered.

physics.soc-ph↗