Minor paper fixes

Author: MarcelRobeer

paper/paper.md

@@ -53,7 +53,7 @@ This fragmentation introduces significant challenges, particularly regarding *re
 `Explabox` transforms opaque *ingestibles* into transparent *digestibles* through four types of *analyses* to enhance explainability and aid fairness, robustness, and security audits.
 
 ## Ingestibles
-Ingestibles provide a unified model/data import interface, where layers abstract away access (\autoref{fig:layers}) to allow optimized processing. `Explabox` uses `instancelib` [@instancelib] for fast model/data encapsulation. The model can be any Python `Callable` containing a regression or (binary and multi-class) classification model.  While this interface is model-agnostic, the current release provides data handling and analysis modules optimized specifically for text-based tasks. `scikit-learn` or `ONNX` models (e.g., `PyTorch`, `TensorFlow`, or `Keras`) import directly with optimizations and automatic input/output interpretation. Data can be automatically downloaded, extracted and loaded. Data inputs include `NumPy`, `Pandas`, or `Hugging Face`, raw files (e.g., HDF5, CSV or TSV), and (compressed) file folders. Data can be subdivided into named splits (e.g., train-test-validation), and instance vectors and tokens can be precomputed and optionally saved for fast inferencing.
+Ingestibles provide a unified model/data import interface, where layers abstract away access (\autoref{fig:layers}) to allow optimized processing. `Explabox` uses `instancelib` [@instancelib] for fast model/data encapsulation. The model can be any Python `Callable` containing a regression or (binary and multi-class) classification model. While this interface is model-agnostic, the current release provides data handling and analysis modules optimized specifically for text-based tasks. `scikit-learn` or `ONNX` models (e.g., `PyTorch`, `TensorFlow`, or `Keras`) import directly with optimizations and automatic input/output interpretation. Data can be automatically downloaded, extracted and loaded. Data inputs include `NumPy`, `Pandas`, `Hugging Face`, raw files (e.g., HDF5, CSV or TSV), and (compressed) file folders. Data can be subdivided into named splits (e.g., train-test-validation), and instance vectors and tokens can be precomputed and optionally saved for fast inferencing.
 
 ![Logical separation of `Explabox` into layers with interfaces.\label{fig:layers}](figure1.png){width=50%}