Jelena Bratulić

David T. Hoffmann, Simon Schrodi, Jelena Bratulić et al.

In this work, we study rapid improvements of the training loss in transformers when being confronted with multi-step decision tasks. We found that transformers struggle to learn the intermediate task and both training and validation loss saturate for hundreds of epochs. When transformers finally learn the intermediate task, they do this rapidly and unexpectedly. We call these abrupt improvements Eureka-moments, since the transformer appears to suddenly learn a previously incomprehensible concept. We designed synthetic tasks to study the problem in detail, but the leaps in performance can be observed also for language modeling and in-context learning (ICL). We suspect that these abrupt transitions are caused by the multi-step nature of these tasks. Indeed, we find connections and show that ways to improve on the synthetic multi-step tasks can be used to improve the training of language modeling and ICL. Using the synthetic data we trace the problem back to the Softmax function in the self-attention block of transformers and show ways to alleviate the problem. These fixes reduce the required number of training steps, lead to higher likelihood to learn the intermediate task, to higher final accuracy and training becomes more robust to hyper-parameters.

Jelena Bratulić, Sudhanshu Mittal, Thomas Brox et al.

The Visual Geometry Grounded Transformer (VGGT) is a 3D foundation model that infers camera geometry and scene structure in a single feed-forward pass. Trained in a supervised, single-step fashion on large datasets, VGGT raises a key question: does it build upon geometric concepts like traditional multi-view methods, or does it rely primarily on learned appearance-based data-driven priors? In this work, we conduct a systematic analysis of VGGT's internal mechanisms to uncover whether geometric understanding emerges within its representations. By probing intermediate features, analyzing attention patterns, and performing interventions, we examine how the model implements its functionality. Our findings reveal that VGGT implicitly performs correspondence matching within its global attention layers and encodes epipolar geometry, despite being trained without explicit geometric constraints. We further investigate VGGT's dependence on its learned data priors. Using spatial input masking and perturbation experiments, we assess its robustness to occlusions, appearance variations, and camera configurations, comparing it with classical multi-stage pipelines. Together, these insights highlight how VGGT internalizes geometric structure while using learned data-driven priors.

Simon Ging, Sebastian Walter, Jelena Bratulić et al.

Training high-quality CLIP models typically requires enormous datasets, which limits the development of domain-specific models -- especially in areas that even the largest CLIP models do not cover well -- and drives up training costs. This poses challenges for scientific research that needs fine-grained control over the training procedure of CLIP models. In this work, we show that by employing smart web search strategies enhanced with knowledge graphs, a robust CLIP model can be trained from scratch with considerably less data. Specifically, we demonstrate that an expert foundation model for living organisms can be built using just 10M images. Moreover, we introduce EntityNet, a dataset comprising 33M images paired with 46M text descriptions, which enables the training of a generic CLIP model in significantly reduced time.

Jelena Bratulić, Sudhanshu Mittal, David T. Hoffmann et al.

Large Language Models (LLMs) exhibit In-Context Learning (ICL), which enables the model to perform new tasks conditioning only on the examples provided in the context without updating the model's weights. While ICL offers fast adaptation across natural language tasks and domains, its emergence is less straightforward for modalities beyond text. In this work, we systematically uncover properties present in LLMs that support the emergence of ICL for autoregressive models and various modalities by promoting the learning of the needed mechanisms for ICL. We identify exact token repetitions in the training data sequences as an important factor for ICL. Such repetitions further improve stability and reduce transiency in ICL performance. Moreover, we emphasise the significance of training task difficulty for the emergence of ICL. Finally, by applying our novel insights on ICL emergence, we unlock ICL capabilities for various visual datasets and a more challenging EEG classification task.