Institution profile

ETH Zurich

Academic institutioneurope · ch
Official website
Research library2,883linked papers
Opportunities0open roles
Selected work

Representative Papers

Realised quantile-based estimation of the integrated variance

Sep 15, 2010

This study addresses the challenges posed by jumps, outliers, and market microstructure noise in high-frequency financial data when estimating realized variance. The authors propose a robust quantile-based estimation method that constructs a quantile-type variance estimator asymptotically immune to finite-activity jumps and outliers, and extend it to noisy high-dimensional settings. Theoretical analysis demonstrates that the proposed estimator consistently recovers the integrated variance at the optimal convergence rate and exhibits favorable asymptotic efficiency. Monte Carlo simulations confirm its pronounced robustness in finite samples, and empirical applications to equity data further validate the practical effectiveness of the approach.

131 citations12 influentialRead paper

Findings of the BabyLM Challenge: Sample-Efficient Pretraining on Developmentally Plausible Corpora

Apr 10, 2025Proceedings of the BabyLM Challenge at the 27th Conference on Computational Natural Language Learning

Large language models (LLMs) suffer from low data efficiency, typically requiring trillion-word corpora for effective pretraining. Method: Inspired by child language acquisition, this work proposes a cognitively grounded, highly efficient pretraining paradigm using only a developmentally appropriate corpus of under 100 million tokens. We systematically demonstrate—contrary to prevailing assumptions—that such small-scale data can surpass trillion-parameter models’ performance when combined with short-sequence training, knowledge distillation, and multi-task evaluation (covering syntactic competence, downstream task transfer, and out-of-distribution generalization); notably, curriculum learning proves ineffective in this low-data regime. Contribution/Results: Leveraging the LTG-BERT architecture, our best-performing model achieves state-of-the-art results across diverse benchmarks, significantly outperforming standard large baselines. The project yields over 30 empirically validated guidelines—identifying both viable strategies and dead ends—for efficient pretraining, thereby establishing a novel paradigm for cognitive modeling and environmentally sustainable (“green”) AI.

105 citations18 influentialRead paper

Developing Generalist Foundation Models from a Multimodal Dataset for 3D Computed Tomography

Mar 26, 2024

Current 3D medical imaging AI is hindered by the scarcity of large-scale, paired multimodal datasets, impeding cross-modal alignment and natural language interaction. To address this, we introduce CT-RATE—the first large-scale, paired 3D chest CT–radiology report dataset (25,692 cases)—and propose CT-CLIP, a contrastive learning framework, and CT-CHAT, a vision-language dialogue model. Our contributions include: (1) the first large-scale, fine-grained alignment between 3D CT volumes and free-text radiology reports; (2) CT-CLIP—a task-agnostic foundational model integrating 3D convolutional networks with Vision Transformers, requiring no downstream fine-tuning; and (3) CT-CHAT—the first open-source, 3D CT–specific conversational model, trained via report-driven QA generation and LLM-CT co-fine-tuning for end-to-end diagnostic interaction. Experiments demonstrate that our unsupervised multi-abnormality detection outperforms fully supervised SOTA methods; cross-modal retrieval enables bidirectional image–text queries; and CT-CHAT, fine-tuned on 2.7M medical QA pairs, surpasses existing multimodal medical assistants.

24 citations3 influentialRead paper

Demystifying Chains, Trees, and Graphs of Thoughts

Jan 25, 2024

Existing structured prompting paradigms—such as Chain-of-Thought (CoT), Tree-of-Thought (ToT), and Graph-of-Thought (GoT)—lack a unified theoretical foundation, suffering from conceptual conflation and an absence of systematic taxonomy. Method: We propose the first comprehensive taxonomy for structured prompting, formally defining the notion of “reasoning topology,” constructing its spatial representation, and unifying CoT, ToT, and GoT through pipeline-based execution analysis, structural modeling, behavioral interpretation, and cross-paradigm empirical comparison. Contribution/Results: (1) We establish the first principled taxonomy for structured-prompt reasoning; (2) we uncover intrinsic relationships between topological structure and both reasoning performance and computational cost; and (3) we provide a theoretically grounded framework and design principles for scalable, interpretable prompt engineering.

21 citations1 influentialRead paper

AtomThink: A Slow Thinking Framework for Multimodal Mathematical Reasoning

Nov 18, 2024arXiv.org

Multimodal large language models (MLLMs) exhibit limited capability in solving complex mathematical reasoning problems due to insufficient granular, stepwise reasoning over multimodal inputs. Method: This paper introduces the “slow-thinking” paradigm, integrating long-chain, atomic-level reasoning into MLLMs via AtomThink—a novel atomic thinking framework comprising (i) an automatic Chain-of-Thought (CoT) annotation engine, (ii) atomic-step fine-tuning, and (iii) a policy-based search method guided by a four-category strategy reward model (PRM). The approach unifies vision–math joint fine-tuning, reinforcement learning–driven search, and interpretable CoT generation. Contribution/Results: We release AtomMATH, a large-scale multimodal mathematical dataset, and propose fine-grained atomic capability evaluation metrics. On MathVista and MathVerse benchmarks, our method achieves relative accuracy improvements of ~50% and ~120%, respectively, significantly enhancing MLLMs’ hierarchical, adaptive reasoning on complex mathematical problems.

18 citations3 influentialRead paper
Recent publications

Latest Papers