Valentina Zantedeschi
Research Scientist
Agentic Harness & Defenses
Valentina is Staff Research Scientist in AI Safety and Time-series Forecasting, Adjunct Professor at Universite’ Laval, and a member of the ELLIS Society. Before joining ServiceNow, Valentina was a post-doctoral researcher at INRIA and University College London (UK), in the context of the INRIA-London Programme, working in Benjamin Guedj’s team on PAC-Bayesian learning. She holds a Ph.D. in Computer Science from University of Lyon (France) with a focus on kernel learning with theoretical guarantees on performance, advised by Marc Sebban and Rémi Emonet. In 2017, she worked as a research intern at IBM Research, Dublin, in Mathieu Sinn’s team, studying and building Deep Learning architectures robust to adversarial examples. The library developed for this research work served as codebase for the release of Adversarial Robustness Toolbox.
Interests
- Time Series Forecasting
- Large Language Models
- Theory of Machine Learning
- Trustworthiness
- Retrieval
- Decission Making
- Safety
- Security
Publications
Bound to Disagree: Generalization Bounds via Certifiable Surrogates.
Conference on Uncertainty in Artificial Intelligence (UAI),
2026.
Dr-CiK: A Testbed for Foresight-Driven Agents.
Workshop at the International Conference of Machine Learning (ICML),
2026.
Hierarchical Retrieval at Scale: Bridging Transparency and Efficiency.
International Conference on Machine Learning (ICML),
2026.
Overcoming the Modality Gap in Context-Aided Forecasting.
International Conference on Machine Learning (ICML),
2026.
Hierarchical Retrieval at Scale: Bridging Transparency and Efficiency.
Workshop at the International Conference of Machine Learning (ICML),
2026.
Overcoming the Modality Gap in Context-Aided Forecasting.
Workshop at the International Conference of Machine Learning (ICML),
2026.
Beyond Naïve Prompting: Strategies for Improved Zero-shot Context-aided Forecasting with LLMs.
Workshop at the Neural Information Processing Systems (NeurIPS),
2025.
Beyond Naïve Prompting: Strategies for Improved Zero-shot Context-aided Forecasting with LLMs.
Conference on Language Modeling Workshops,
2025.
Context is Key: A Benchmark for Forecasting with Essential Textual Information.
International Conference on Machine Learning (ICML),
2025.
Sample compression unleashed: New generalization bounds for real valued losses.
International Conference on Artificial Intelligence and Statistics (AISTATS),
2025.
Learning to Defer for Causal Discovery with Imperfect Experts.
Workshop at the International Conference of Learning Representation (ICLR),
2025.
InsightBench: Evaluating Business Analytics Agents Through Multi-Step Insight Generation.
International Conference of Learning Representations (ICLR),
2025.
Context is Key: A Benchmark for Forecasting with Essential Textual Information.
Workshop at the Neural Information Processing Systems (NeurIPS),
2024.
RepLiQA: A Question-Answering Dataset for Benchmarking LLMs on Unseen Reference Content.
NeurIPS Datasets and Benchmarks Track (NeurIPS Datasets),
2024.
Sample compression unleashed: New generalization bounds for real valued losses.
Workshop at the Neural Information Processing Systems (NeurIPS),
2024.
XC-Cache: Cross-Attending to Cached Context for Efficient LLM Inference.
Workshop at the Neural Information Processing Systems (NeurIPS),
2024.
Context is Key: A Benchmark for Forecasting with Essential Textual Information.
Foundation Models for Time Series,
2024.
XC-Cache: Cross-Attending to Cached Context for Efficient LLM Inference.
Conference on Empirical Methods in Natural Language Processing (EMNLP),
2024.
Context is Key: A Benchmark for Forecasting with Essential Textual Information.
Montreal AI Symposium (MAIS),
2024.
TACTIS-2: Better, Faster, Simpler Attentional Copulas for Multivariate Time Series.
Montreal AI Symposium (MAIS),
2024.
Performance Control in Early Exiting to Deploy Large Models at the Same Cost of Smaller Ones.
Workshop at the International Conference of Machine Learning (ICML),
2024.
TACTIS-2: Better, Faster, Simpler Attentional Copulas for Multivariate Time Series.
International Conference of Learning Representations (ICLR),
2024.
Generalization bounds with arbitrary complexity measures.
International Conference on Artificial Intelligence and Statistics (AISTATS),
2024.
Capture the Flag: Uncovering Data Insights with Large Language Models.
Workshop at the Neural Information Processing Systems (NeurIPS),
2023.
Causal Discovery with Language Models as Imperfect Experts.
Workshop on Structured Probabilistic Inference & Generative Modeling (ICML),
2023.
Regions of Reliability in the Evaluation of Multivariate Probabilistic Forecasts.
International Conference on Machine Learning (ICML),
2023.
DAG Learning on the Permutahedron.
International Conference of Learning Representations (ICLR),
2023.
Learning Discrete Directed Acyclic Graphs via Backpropagation.
Workshop at the Neural Information Processing Systems (NeurIPS),
2022.
On Margins and Generalisation for Voting Classifiers.
Conference on Neural Information Processing Systems (NeurIPS),
2022.
RaVAEn: Unsupervised Change Detection of Extreme Events Using ML On-Board Satellites.
Nature Scientific Reports,
2022.