Martin A. Giese

Jana Lang, Martin Giese, Winfried Ilg et al.

Since neural networks play an increasingly important role in critical sectors, explaining network predictions has become a key research topic. Counterfactual explanations can help to understand why classifier models decide for particular class assignments and, moreover, how the respective input samples would have to be modified such that the class prediction changes. Previous approaches mainly focus on image and tabular data. In this work we propose SPARCE, a generative adversarial network (GAN) architecture that generates SPARse Counterfactual Explanations for multivariate time series. Our approach provides a custom sparsity layer and regularizes the counterfactual loss function in terms of similarity, sparsity, and smoothness of trajectories. We evaluate our approach on real-world human motion datasets as well as a synthetic time series interpretability benchmark. Although we make significantly sparser modifications than other approaches, we achieve comparable or better performance on all metrics. Moreover, we demonstrate that our approach predominantly modifies salient time steps and features, leaving non-salient inputs untouched.

Yuanwei Qu, Michel Perrin, Anita Torabi et al.

Geological modeling currently uses various computer-based applications. Data harmonization at the semantic level by means of ontologies is essential for making these applications interoperable. Since geo-modeling is currently part of multidisciplinary projects, semantic harmonization is required to model not only geological knowledge but also to integrate other domain knowledge at a general level. For this reason, the domain ontologies used for describing geological knowledge must be based on a sound ontology background to ensure the described geological knowledge is integratable. This paper presents a domain ontology: GeoFault, resting on the Basic Formal Ontology BFO (Arp et al., 2015) and the GeoCore ontology (Garcia et al., 2020). It models the knowledge related to geological faults. Faults are essential to various industries but are complex to model. They can be described as thin deformed rock volumes or as spatial arrangements resulting from the different displacements of geological blocks. At a broader scale, faults are currently described as mere surfaces, which are the components of complex fault arrays. The reference to the BFO and GeoCore package allows assigning these various fault elements to define ontology classes and their logical linkage within a consistent ontology framework. The GeoFault ontology covers the core knowledge of faults 'strico sensu,' excluding ductile shear deformations. This considered vocabulary is essentially descriptive and related to regional to outcrop scales, excluding microscopic, orogenic, and tectonic plate structures. The ontology is molded in OWL 2, validated by competency questions with two use cases, and tested using an in-house ontology-driven data entry application. The work of GeoFault provides a solid framework for disambiguating fault knowledge and a foundation of fault data integration for the applications and the users.

Prerana Kumar, Martin A. Giese

Action recognition is a fundamental ability for social species. Yet, its underlying computations are not well understood. Classical psychophysical studies using simplified stimuli have shown that humans can perceive body motion even under degradation of relevant shape cues. Recent work using real-world action videos and their appearance-free counterparts (that preserve motion but lack static shape cues) included explicit training of humans and models on the appearance-free videos. Whether humans and vision models generalize in a zero-shot manner to appearance-free transformations of real-world action videos is not yet known. To measure this generalization in humans, we conducted a laboratory-based psychophysics experiment. 22 participants were trained to recognize five action categories using naturalistic videos (UCF5 dataset), and tested zero-shot on two types of appearance-free transformations: (i) dense-noise motion videos from an existing dataset (AFD5) and (ii) random-dot appearance-free videos. We find that participants recognize actions in both types of appearance-free videos well above chance, albeit with reduced accuracy compared to naturalistic videos. To model this behavior, we developed a two-pathway 3D CNN-based model combining an RGB (form) stream and an optical flow (motion) stream, including a coherence-gating mechanism inspired by Gestalt common-fate grouping. Our model generalizes to both appearance-free datasets and outperforms contemporary video classification models, narrowing the gap to human performance. We find that the motion pathway is critical for generalization to appearance-free videos, while the form pathway improves performance on naturalistic videos. Our findings highlight the importance of motion-based representations for generalization to appearance-free videos, and support the use of multi-stream architectures to model video-based action recognition.

Michael Stettler, Alexander Lappe, Nick Taubert et al.

People can innately recognize human facial expressions in unnatural forms, such as when depicted on the unusual faces drawn in cartoons or when applied to an animal's features. However, current machine learning algorithms struggle with out-of-domain transfer in facial expression recognition (FER). We propose a biologically-inspired mechanism for such transfer learning, which is based on norm-referenced encoding, where patterns are encoded in terms of difference vectors relative to a domain-specific reference vector. By incorporating domain-specific reference frames, we demonstrate high data efficiency in transfer learning across multiple domains. Our proposed architecture provides an explanation for how the human brain might innately recognize facial expressions on varying head shapes (humans, monkeys, and cartoon avatars) without extensive training. Norm-referenced encoding also allows the intensity of the expression to be read out directly from neural unit activity, similar to face-selective neurons in the brain. Our model achieves a classification accuracy of 92.15\% on the FERG dataset with extreme data efficiency. We train our proposed mechanism with only 12 images, including a single image of each class (facial expression) and one image per domain (avatar). In comparison, the authors of the FERG dataset achieved a classification accuracy of 89.02\% with their FaceExpr model, which was trained on 43,000 images.

Alexander Lappe, Martin A. Giese

Recent work has shown that the attention maps of the widely popular DINOv2 model exhibit artifacts, which hurt both model interpretability and performance on dense image tasks. These artifacts emerge due to the model repurposing patch tokens with redundant local information for the storage of global image information. To address this problem, additional register tokens have been incorporated in which the model can store such information instead. We carefully examine the influence of these register tokens on the relationship between global and local image features, showing that while register tokens yield cleaner attention maps, these maps do not accurately reflect the integration of local image information in large models. Instead, global information is dominated by information extracted from register tokens, leading to a disconnect between local and global features. Inspired by these findings, we show that the [CLS] token itself leads to a very similar phenomenon in models without explicit register tokens. Our work shows that care must be taken when interpreting attention maps of large ViTs. Further, by clearly attributing the faulty behavior to register and [CLS] tokens, we show a path towards more interpretable vision models.

Alexander Lappe, Anna Bognár, Ghazaleh Ghamkhari Nejad et al.

High-level visual brain regions contain subareas in which neurons appear to respond more strongly to examples of a particular semantic category, like faces or bodies, rather than objects. However, recent work has shown that while this finding holds on average, some out-of-category stimuli also activate neurons in these regions. This may be due to visual features common among the preferred class also being present in other images. Here, we propose a deep-learning-based approach for visualizing these features. For each neuron, we identify relevant visual features driving its selectivity by modelling responses to images based on latent activations of a deep neural network. Given an out-of-category image which strongly activates the neuron, our method first identifies a reference image from the preferred category yielding a similar feature activation pattern. We then backpropagate latent activations of both images to the pixel level, while enhancing the identified shared dimensions and attenuating non-shared features. The procedure highlights image regions containing shared features driving responses of the model neuron. We apply the algorithm to novel recordings from body-selective regions in macaque IT cortex in order to understand why some images of objects excite these neurons. Visualizations reveal object parts which resemble parts of a macaque body, shedding light on neural preference of these objects.

Jesse St. Amand, Leonardo Gizzi, Martin A. Giese

We present the Gaussian process dynamical mixture model (GPDMM) and show its utility in single-example learning of human motion data. The Gaussian process dynamical model (GPDM) is a form of the Gaussian process latent variable model (GPLVM), but optimized with a hidden Markov model dynamical prior. The GPDMM combines multiple GPDMs in a probabilistic mixture-of-experts framework, utilizing embedded geometric features to allow for diverse sequences to be encoded in a single latent space, enabling the categorization and generation of each sequence class. GPDMs and our mixture model are particularly advantageous in addressing the challenges of modeling human movement in scenarios where data is limited and model interpretability is vital, such as in patient-specific medical applications like prosthesis control. We score the GPDMM on classification accuracy and generative ability in single-example learning, showcase model variations, and benchmark it against LSTMs, VAEs, and transformers.

Jana Lang, Martin A. Giese, Matthis Synofzik et al.

Motor disturbances can affect the interaction with dynamic objects, such as catching a ball. A classification of clinical catching trials might give insight into the existence of pathological alterations in the relation of arm and ball movements. Accurate, but also early decisions are required to classify a catching attempt before the catcher's first ball contact. To obtain clinically valuable results, a significant decision confidence of at least 75% is required. Hence, three competing objectives have to be optimized at the same time: accuracy, earliness and decision-making confidence. Here we propose a coupled classification and prediction approach for early time series classification: a predictive, generative recurrent neural network (RNN) forecasts the next data points of ball trajectories based on already available observations; a discriminative RNN continuously generates classification guesses based on the available data points and the unrolled sequence predictions. We compare our approach, which we refer to as predictive sequential classification (PSC), to state-of-the-art sequence learners, including various RNN and temporal convolutional network (TCN) architectures. On this hard real-world task we can consistently demonstrate the superiority of PSC over all other models in terms of accuracy and confidence with respect to earliness of recognition. Specifically, PSC is able to confidently classify the success of catching trials as early as 123 milliseconds before the first ball contact. We conclude that PSC is a promising approach for early time series classification, when accurate and confident decisions are required.

Alessandro Salatiello, Martin A. Giese

State-of-the-art motorized hand prostheses are endowed with actuators able to provide independent and proportional control of as many as six degrees of freedom (DOFs). The control signals are derived from residual electromyographic (EMG) activity, recorded concurrently from relevant forearm muscles. Nevertheless, the functional mapping between forearm EMG activity and hand kinematics is only known with limited accuracy. Therefore, no robust method exists for the reliable computation of control signals for the independent and proportional actuation of more than two DOFs. A common approach to deal with this limitation is to pre-program the prostheses for the execution of a restricted number of behaviors (e.g., pinching, grasping, and wrist rotation) that are activated by the detection of specific EMG activation patterns. However, this approach severely limits the range of activities users can perform with the prostheses during their daily living. In this work, we introduce a novel method, based on a long short-term memory (LSTM) network, to continuously map forearm EMG activity onto hand kinematics. Critically, unlike previous work, which often focuses on simple and highly controlled motor tasks, we tested our method on a dataset of activities of daily living (ADLs): the KIN-MUS UJI dataset. To the best of our knowledge, ours is the first reported work on the prediction of hand kinematics that uses this challenging dataset. Remarkably, we show that our network is able to generalize to novel untrained ADLs. Our results suggest that the presented method is suitable for the generation of control signals for the independent and proportional actuation of the multiple DOFs of state-of-the-art hand prostheses.

Alessandro Salatiello, Martin A. Giese

Recurrent Neural Networks (RNNs) are popular models of brain function. The typical training strategy is to adjust their input-output behavior so that it matches that of the biological circuit of interest. Even though this strategy ensures that the biological and artificial networks perform the same computational task, it does not guarantee that their internal activity dynamics match. This suggests that the trained RNNs might end up performing the task employing a different internal computational mechanism, which would make them a suboptimal model of the biological circuit. In this work, we introduce a novel training strategy that allows learning not only the input-output behavior of an RNN but also its internal network dynamics, based on sparse neural recordings. We test the proposed method by training an RNN to simultaneously reproduce internal dynamics and output signals of a physiologically-inspired neural model. Specifically, this model generates the multiphasic muscle-like activity patterns typically observed during the execution of reaching movements, based on the oscillatory activation patterns concurrently observed in the motor cortex. Remarkably, we show that the reproduction of the internal dynamics is successful even when the training algorithm relies on the activities of a small subset of neurons sampled from the biological network. Furthermore, we show that training the RNNs with this method significantly improves their generalization performance. Overall, our results suggest that the proposed method is suitable for building powerful functional RNN models, which automatically capture important computational properties of the biological circuit of interest from sparse neural recordings.