Confidence Estimation for Financial Vision-Language Models in Chart and Document Understanding

LVLMs are increasingly used to read financial charts, tables, and documents, where a single misread figure can move a decision and the most authoritative-looking answer is sometimes one the model produced without reading the exhibit. The operational question is therefore trust, not accuracy: which answers can be acted on, and which escalated to a reviewer. We evaluate seven confidence estimators, three inference-only and four trained internal probes, across five open-weight LVLMs and four conditions from three financial visual question-answering benchmarks, one bilingual; every probe is trained only on natural images and applied to finance without adaptation, so the results measure out-of-distribution transfer. Three findings hold. First, the scarce property is calibration, not ranking: the inference baselines rank correct above incorrect answers competitively but are badly overconfident, calibration error far above what a threshold can tolerate, and only the trained probes produce a thresholdable score. Second, reliability is structured rather than global, along two axes a practitioner can read directly: the best estimator shifts with both model and task, none leading more than eight of twenty (model, condition) cells, and a controlled bilingual contrast exposes an apparent language robustness as a composition artifact that dissolves once models are read one at a time. Third, cast as deferral under an error budget, how much can be safely automated is set first by the model's competence and only narrowed by its confidence, so deferral clears a real share of the easiest condition and almost none of the hardest, near zero at a strict 5% budget. Two trained probes carry the calibration a deferral policy needs, and among them only the grounding-aware one lowers its confidence on answers a model gives without using the figure, separating detected non-grounding from a fluent guess.

Model Confidence Under Answer-Preserving Attacks: An Informativeness-Manipulability Frontier

Deployed vision-language systems often gate their answers on confidence, making confidence robustness relevant to oversight. We study confidence readouts under white-box, image-only attacks constrained to preserve the generated answer byte-identically. Under a reachability assumption, an unmovable readout cannot outperform the answer-string accuracy prior, whose pooled value is 0.617. Independently of that assumption, a uniform amplitude certificate below a measurable threshold guarantees adversarial discrimination above the same floor. Across four vision-language models, three visual question answering benchmarks, five deployed confidence channels and two defense estimators, direct or surrogate-aimed attacks produce itemwise feasible perturbations that refute this uniform certificate in all 84 estimator-by-cell combinations. Coordinated correctness-label-aware attacks drive adversarial discrimination to or below the answer-string floor in all sixty deployed-channel cells, including all fifty-nine that begin above it. Hidden-state interventions and an open-ended text-model activation-space replication show that comparable confidence movement can be induced at the representation level rather than only through adversarial images. None of four tested defense families establishes a robust alternative under the specific evaluation applied to it. In a confidence-gated simulation, a coordinated token-probability attack transferred to a hidden-state gate causes up to 84.8% of previously rejected wrong answers to become accepted. After reweighting to each benchmark's natural correctness prevalence, accepted accuracy falls below the no-gate baseline in eight of twelve cells under transfer and all twelve under a direct gate-aimed attack. Under the studied threat model and budget, confidence is therefore an integrity-sensitive rather than intrinsically robust oversight signal.

Calibrated Triage, Not Autonomy: Confidence Estimation for Medical Vision-Language Models

A vision-language model can answer a question about a medical image fluently and confidently while barely using the image, leaning instead on language priors. In medicine this is the failure that matters most, because the answer looks trustworthy and is not, and the only protection is a confidence score reliable enough to tell the system when to abstain. We ask a deployment question rather than an accuracy one: how much imaging work a model can safely handle alone, and which confidence signal makes that possible. We evaluate seven confidence estimators across five open-weight LVLMs and three medical visual-question-answering datasets spanning broad clinical imaging, radiology, and pathology, with every probe trained only on natural images and applied without adaptation. Recast as bounded selective prediction (automate a case only when confidence clears a threshold, defer the rest), the comparison is cautionary. The standard metrics are poor guides: discrimination barely separates the methods, and the weak calibration of a cheap self-report is cheaply removed by off-domain temperature scaling without changing deployable yield. What distinguishes a usable estimator is the high-confidence region a clinician acts on: the weakest baselines are confidently wrong on 41 to 45 percent of their errors against 1 to 4 percent for the best probe, and no estimator is reliably best across domains or models. Safe handoff is governed at two levels: base-model competence sets a ceiling, so a well-calibrated score recovers roughly a third of radiology cases at a 20 percent error tolerance but almost none of pathology; the confidence layer then decides how much of that ceiling is reachable. The usable role today is calibrated triage, not autonomy: automate the cases a calibrated score marks safe, route the rest to a clinician. We release all outputs, correctness judgments, and confidence scores, with code.

Grounded or Guessing? LVLM Confidence Estimation via Blind-Image Contrastive Ranking

Large vision-language models suffer from visual ungroundedness: they can produce a fluent, confident, and even correct response driven entirely by language priors, with the image contributing nothing to the prediction. Existing confidence estimation methods cannot detect this, as they observe model behavior under normal inference with no mechanism to determine whether a prediction was shaped by the image or by text alone. We introduce BICR (Blind-Image Contrastive Ranking), a model-agnostic confidence estimation framework that makes this contrast explicit during training by extracting hidden states from a frozen LVLM twice: once with the real image-question pair, and once with the image blacked out while the question is held fixed. A lightweight probe is trained on the real-image hidden state and regularized by a ranking loss that penalizes higher confidence on the blacked-out view, teaching it to treat visual grounding as a signal of reliability at zero additional inference cost. Evaluated across five modern LVLMs and seven baselines on a benchmark covering visual question answering, object hallucination detection, medical imaging, and financial document understanding, BICR achieves the best cross-LVLM average on both calibration and discrimination simultaneously, with statistically significant discrimination gains robust to cluster-aware analysis at 4-18x fewer parameters than the strongest probing baseline.

What Language Models Know But Don't Say: Non-Generative Prior Extraction for Generalization

In domains like medicine and finance, large-scale labeled data is costly and often unavailable, leading to models trained on small datasets that struggle to generalize to real-world populations. Large language models contain extensive knowledge from years of research across these domains. We propose LoID (Logit-Informed Distributions), a deterministic method for extracting informative prior distributions for Bayesian logistic regression by directly accessing their token-level predictions. Rather than relying on generated text, we probe the model's confidence in opposing semantic directions (positive vs. negative impact) through carefully constructed sentences. By measuring how consistently the LLM favors one direction across diverse phrasings, we extract the strength and reliability of the model's belief about each feature's influence. We evaluate LoID on ten real-world tabular datasets under synthetic out-of-distribution (OOD) settings characterized by covariate shift, where the training data represents only a subset of the population. We compare our approach against (1) standard uninformative priors, (2) AutoElicit, a recent method that prompts LLMs to generate priors via text completions, (3) LLMProcesses, a method that uses LLMs to generate numerical predictions through in-context learning and (4) an oracle-style upper bound derived from fitting logistic regression on the full dataset. We assess performance using Area Under the Curve (AUC). Across datasets, LoID significantly improves performance over logistic regression trained on OOD data, recovering up to \textbf{59\%} of the performance gap relative to the oracle model. LoID outperforms AutoElicit and LLMProcessesc on 8 out of 10 datasets, while providing a reproducible and computationally efficient mechanism for integrating LLM knowledge into Bayesian inference.

How Reliable are Confidence Estimators for Large Reasoning Models? A Systematic Benchmark on High-Stakes Domains

The miscalibration of Large Reasoning Models (LRMs) undermines their reliability in high-stakes domains, necessitating methods to accurately estimate the confidence of their long-form, multi-step outputs. To address this gap, we introduce the Reasoning Model Confidence estimation Benchmark (RMCB), a public resource of 347,496 reasoning traces from six popular LRMs across different architectural families. The benchmark is constructed from a diverse suite of datasets spanning high-stakes domains, including clinical, financial, legal, and mathematical reasoning, alongside complex general reasoning benchmarks, with correctness annotations provided for all samples. Using RMCB, we conduct a large-scale empirical evaluation of over ten distinct representation-based methods, spanning sequential, graph-based, and text-based architectures. Our central finding is a persistent trade-off between discrimination (AUROC) and calibration (ECE): text-based encoders achieve the best AUROC (0.672), while structurally-aware models yield the best ECE (0.148), with no single method dominating both. Furthermore, we find that increased architectural complexity does not reliably outperform simpler sequential baselines, suggesting a performance ceiling for methods relying solely on chunk-level hidden states. This work provides the most comprehensive benchmark for this task to date, establishing rigorous baselines and demonstrating the limitations of current representation-based paradigms.

Extracting Overlapping Microservices from Monolithic Code via Deep Semantic Embeddings and Graph Neural Network-Based Soft Clustering

Modern software systems are increasingly shifting from monolithic architectures to microservices to enhance scalability, maintainability, and deployment flexibility. Existing microservice extraction methods typically rely on hard clustering, assigning each software component to a single microservice. This approach often increases inter-service coupling and reduces intra-service cohesion. We propose Mo2oM (Monolithic to Overlapping Microservices), a framework that formulates microservice extraction as a soft clustering problem, allowing components to belong probabilistically to multiple microservices. This approach is inspired by expert-driven decompositions, where practitioners intentionally replicate certain software components across services to reduce communication overhead. Mo2oM combines deep semantic embeddings with structural dependencies extracted from methodcall graphs to capture both functional and architectural relationships. A graph neural network-based soft clustering algorithm then generates the final set of microservices. We evaluate Mo2oM on four open-source monolithic benchmarks and compare it against eight state-of-the-art baselines. Our results demonstrate that Mo2oM achieves improvements of up to 40.97% in structural modularity (balancing cohesion and coupling), 58% in inter-service call percentage (communication overhead), 26.16% in interface number (modularity and decoupling), and 38.96% in non-extreme distribution (service size balance) across all benchmarks.

Calibrating LLM Confidence by Probing Perturbed Representation Stability

Miscalibration in Large Language Models (LLMs) undermines their reliability, highlighting the need for accurate confidence estimation. We introduce CCPS (Calibrating LLM Confidence by Probing Perturbed Representation Stability), a novel method analyzing internal representational stability in LLMs. CCPS applies targeted adversarial perturbations to final hidden states, extracts features reflecting the model's response to these perturbations, and uses a lightweight classifier to predict answer correctness. CCPS was evaluated on LLMs from 8B to 32B parameters (covering Llama, Qwen, and Mistral architectures) using MMLU and MMLU-Pro benchmarks in both multiple-choice and open-ended formats. Our results show that CCPS significantly outperforms current approaches. Across four LLMs and three MMLU variants, CCPS reduces Expected Calibration Error by approximately 55% and Brier score by 21%, while increasing accuracy by 5 percentage points, Area Under the Precision-Recall Curve by 4 percentage points, and Area Under the Receiver Operating Characteristic Curve by 6 percentage points, all relative to the strongest prior method. CCPS delivers an efficient, broadly applicable, and more accurate solution for estimating LLM confidence, thereby improving their trustworthiness.

MoENAS: Mixture-of-Expert based Neural Architecture Search for jointly Accurate, Fair, and Robust Edge Deep Neural Networks

There has been a surge in optimizing edge Deep Neural Networks (DNNs) for accuracy and efficiency using traditional optimization techniques such as pruning, and more recently, employing automatic design methodologies. However, the focus of these design techniques has often overlooked critical metrics such as fairness, robustness, and generalization. As a result, when evaluating SOTA edge DNNs' performance in image classification using the FACET dataset, we found that they exhibit significant accuracy disparities (14.09%) across 10 different skin tones, alongside issues of non-robustness and poor generalizability. In response to these observations, we introduce Mixture-of-Experts-based Neural Architecture Search (MoENAS), an automatic design technique that navigates through a space of mixture of experts to discover accurate, fair, robust, and general edge DNNs. MoENAS improves the accuracy by 4.02% compared to SOTA edge DNNs and reduces the skin tone accuracy disparities from 14.09% to 5.60%, while enhancing robustness by 3.80% and minimizing overfitting to 0.21%, all while keeping model size close to state-of-the-art models average size (+0.4M). With these improvements, MoENAS establishes a new benchmark for edge DNN design, paving the way for the development of more inclusive and robust edge DNNs.

Online Continual Learning: A Systematic Literature Review of Approaches, Challenges, and Benchmarks

Online Continual Learning (OCL) is a critical area in machine learning, focusing on enabling models to adapt to evolving data streams in real-time while addressing challenges such as catastrophic forgetting and the stability-plasticity trade-off. This study conducts the first comprehensive Systematic Literature Review (SLR) on OCL, analyzing 81 approaches, extracting over 1,000 features (specific tasks addressed by these approaches), and identifying more than 500 components (sub-models within approaches, including algorithms and tools). We also review 83 datasets spanning applications like image classification, object detection, and multimodal vision-language tasks. Our findings highlight key challenges, including reducing computational overhead, developing domain-agnostic solutions, and improving scalability in resource-constrained environments. Furthermore, we identify promising directions for future research, such as leveraging self-supervised learning for multimodal and sequential data, designing adaptive memory mechanisms that integrate sparse retrieval and generative replay, and creating efficient frameworks for real-world applications with noisy or evolving task boundaries. By providing a rigorous and structured synthesis of the current state of OCL, this review offers a valuable resource for advancing this field and addressing its critical challenges and opportunities. The complete SLR methodology steps and extracted data are publicly available through the provided link: https://github.com/kiyan-rezaee/ Systematic-Literature-Review-on-Online-Continual-Learning