Selected work

Representative Papers

Herding End-to-End Autonomous Driving via Neuro-Symbolic Safety Guards

Aug 11, 2026

This work addresses the safety and interpretability limitations of end-to-end autonomous driving systems, which often violate traffic rules due to reliance on statistical data patterns and lack verifiable guarantees. The authors propose a lightweight neuro-symbolic safety guard that enforces explicit traffic regulations in real time—without retraining the underlying model or adding new learning components—by validating control commands prior to execution and substituting them with the nearest feasible safe action when necessary. Integrated seamlessly into the TransFuser v6 architecture, this mechanism ensures every intervention is traceable to a specific rule. Experiments on the Fail2Drive and Bench2Drive long-tail benchmarks demonstrate a 53% reduction in severe collisions and a 15% improvement in task success rate, all while preserving the original driving performance.

0 citationsRead paper

Structured Phonological Representations for Audio-Articulatory rtMRI Speech Classification

Aug 10, 2026

Mapping real-time magnetic resonance imaging (rtMRI)–derived articulatory movements to phonological categories remains challenging. This work proposes a multimodal modeling approach that, for the first time, transfers audio representations trained with phonological feature supervision—based on PhonoQ, WavLM-large, and HuBERT-large—into purely articulatory modeling, yielding performance gains even during audio-free inference. By fusing synchronized audio and articulatory contours, the model significantly improves macro-F1 scores for phonological classification tasks (e.g., manner and place of articulation, voicing, vowel height, and backness) under both unseen-speech and unseen-speaker conditions, while also enhancing fine-grained accuracy across 39 phoneme classes. Furthermore, the approach reveals interpretable patterns of articulatory variation.

0 citationsRead paper

When Repository Labels Are Not Image-Level Truth: A Supervision Auditing Framework for Chest Radiograph AI

Aug 10, 2026

This study addresses the widespread issue of label inaccuracies in public chest X-ray datasets, where annotations derived from radiology reports often misrepresent actual pathological findings, thereby compromising model training with unreliable supervisory signals. To mitigate this, the authors propose the Repository Supervision Auditing (RSA) framework, which employs expert image-level annotations to audit label consistency prior to model development, identify systematic biases, and construct a verified evaluation cohort. Their analysis reveals severe discrepancies between report-based labels and radiological evidence: only 1% of cases with expert-confirmed cardiomegaly were correctly labeled, while nearly half were erroneously annotated as “no abnormality.” A DenseNet121 model trained on the corrected cohort achieved a test ROC-AUC of 0.853, demonstrating that supervision auditing is a critical prerequisite for robust medical imaging AI development.

0 citationsRead paper

An Inflation Model for the Colombian Case. 2001 2025

Mar 27, 2026

This study investigates inflation dynamics and the effectiveness of monetary policy in Colombia under an inflation-targeting regime with a floating exchange rate. Utilizing data from 2001 to 2025, the authors develop a structural model that integrates the inflation-targeting framework with a structural Phillips curve incorporating the output gap and inflation expectations, estimating it via generalized method of moments (GMM). The empirical analysis confirms the existence of a Phillips curve conditional on inflation expectations and reveals that while the central bank’s policy stance is directionally appropriate, its response is both delayed and insufficient in magnitude to promptly anchor inflation within the target range. These findings offer new empirical evidence on the importance of timeliness and credibility in monetary policy implementation, particularly for emerging market economies.

0 citationsRead paper

Efficient reduction of stellar contamination and noise in planetary transmission spectra using neural networks

Feb 10, 2026

Stellar surface inhomogeneities, such as spots and faculae, can severely contaminate exoplanetary transmission spectra, leading to significant biases in atmospheric parameter retrieval. This work proposes an unsupervised preprocessing method based on a denoising autoencoder, which—unlike conventional joint-fitting approaches—simultaneously removes stellar contamination and instrumental noise without incurring high computational costs. Trained on synthetic data, the method employs χ² statistics to assess reconstruction fidelity and integrates seamlessly into standard atmospheric retrieval pipelines. Experiments demonstrate that it effectively preserves key molecular features even at low signal-to-noise ratios, yielding retrieval biases substantially lower than those of uncorrected baselines and achieving accuracy comparable to joint fitting, while reducing computational time by a factor of 3–6. The approach is applicable to both rocky and gaseous exoplanets.

0 citationsRead paper

Recent publications

Latest Papers

Herding End-to-End Autonomous Driving via Neuro-Symbolic Safety Guards

Aug 11, 2026

This work addresses the safety and interpretability limitations of end-to-end autonomous driving systems, which often violate traffic rules due to reliance on statistical data patterns and lack verifiable guarantees. The authors propose a lightweight neuro-symbolic safety guard that enforces explicit traffic regulations in real time—without retraining the underlying model or adding new learning components—by validating control commands prior to execution and substituting them with the nearest feasible safe action when necessary. Integrated seamlessly into the TransFuser v6 architecture, this mechanism ensures every intervention is traceable to a specific rule. Experiments on the Fail2Drive and Bench2Drive long-tail benchmarks demonstrate a 53% reduction in severe collisions and a 15% improvement in task success rate, all while preserving the original driving performance.

0 citationsRead paper

Structured Phonological Representations for Audio-Articulatory rtMRI Speech Classification

Aug 10, 2026

Mapping real-time magnetic resonance imaging (rtMRI)–derived articulatory movements to phonological categories remains challenging. This work proposes a multimodal modeling approach that, for the first time, transfers audio representations trained with phonological feature supervision—based on PhonoQ, WavLM-large, and HuBERT-large—into purely articulatory modeling, yielding performance gains even during audio-free inference. By fusing synchronized audio and articulatory contours, the model significantly improves macro-F1 scores for phonological classification tasks (e.g., manner and place of articulation, voicing, vowel height, and backness) under both unseen-speech and unseen-speaker conditions, while also enhancing fine-grained accuracy across 39 phoneme classes. Furthermore, the approach reveals interpretable patterns of articulatory variation.

0 citationsRead paper

When Repository Labels Are Not Image-Level Truth: A Supervision Auditing Framework for Chest Radiograph AI

Aug 10, 2026

This study addresses the widespread issue of label inaccuracies in public chest X-ray datasets, where annotations derived from radiology reports often misrepresent actual pathological findings, thereby compromising model training with unreliable supervisory signals. To mitigate this, the authors propose the Repository Supervision Auditing (RSA) framework, which employs expert image-level annotations to audit label consistency prior to model development, identify systematic biases, and construct a verified evaluation cohort. Their analysis reveals severe discrepancies between report-based labels and radiological evidence: only 1% of cases with expert-confirmed cardiomegaly were correctly labeled, while nearly half were erroneously annotated as “no abnormality.” A DenseNet121 model trained on the corrected cohort achieved a test ROC-AUC of 0.853, demonstrating that supervision auditing is a critical prerequisite for robust medical imaging AI development.

0 citationsRead paper

An Inflation Model for the Colombian Case. 2001 2025

Mar 27, 2026

This study investigates inflation dynamics and the effectiveness of monetary policy in Colombia under an inflation-targeting regime with a floating exchange rate. Utilizing data from 2001 to 2025, the authors develop a structural model that integrates the inflation-targeting framework with a structural Phillips curve incorporating the output gap and inflation expectations, estimating it via generalized method of moments (GMM). The empirical analysis confirms the existence of a Phillips curve conditional on inflation expectations and reveals that while the central bank’s policy stance is directionally appropriate, its response is both delayed and insufficient in magnitude to promptly anchor inflation within the target range. These findings offer new empirical evidence on the importance of timeliness and credibility in monetary policy implementation, particularly for emerging market economies.

0 citationsRead paper

Efficient reduction of stellar contamination and noise in planetary transmission spectra using neural networks

Feb 10, 2026

Stellar surface inhomogeneities, such as spots and faculae, can severely contaminate exoplanetary transmission spectra, leading to significant biases in atmospheric parameter retrieval. This work proposes an unsupervised preprocessing method based on a denoising autoencoder, which—unlike conventional joint-fitting approaches—simultaneously removes stellar contamination and instrumental noise without incurring high computational costs. Trained on synthetic data, the method employs χ² statistics to assess reconstruction fidelity and integrates seamlessly into standard atmospheric retrieval pipelines. Experiments demonstrate that it effectively preserves key molecular features even at low signal-to-noise ratios, yielding retrieval biases substantially lower than those of uncorrected baselines and achieving accuracy comparable to joint fitting, while reducing computational time by a factor of 3–6. The approach is applicable to both rocky and gaseous exoplanets.

0 citationsRead paper