Abstract 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 mechanisms needed 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. We analyse in detail the training dynamics of such data sequences and explain how token repetitions enhance the ICL learning mechanisms. Moreover, we emphasise the importance of the training task difficulty for the emergence of ICL. Finally, by applying our novel insights on ICL emergence, we unlock ICL capabilities across various visual datasets used for few-shot classification, and confirm the generalisability of our insights to much harder real-world examples of large-scale object classification, and a more challenging EEG classification task. Code is available at https://github.com/jelenab98/unlocking_icl

While tabular classification has traditionally relied on from-scratch training, a recent breakthrough called prior-data fitted networks (PFNs) challenges this approach. Similar to large language models, PFNs make use of pretraining and in-context learning to achieve strong performance on new tasks in a single forward pass. However, current PFNs have limitations that prohibit their widespread adoption. Notably, TabPFN achieves very strong performance on small tabular datasets but is not designed to make predictions for datasets of size larger than 1000. In this work, we overcome these limitations and substantially improve the performance of PFNs via context optimization. We introduce TuneTables, a parameter-efficient fine-tuning strategy for PFNs that compresses large datasets into a smaller learned context. We conduct extensive experiments on nineteen algorithms over 98 datasets and find that TuneTables achieves the best performance on average, outperforming boosted trees such as CatBoost, while optimizing fewer than 5\% of TabPFN's parameters. Furthermore, we show that TuneTables can be used as an interpretability tool and can even be used to mitigate biases by optimizing a fairness objective.

Deep generative networks trained via maximum likelihood on a natural image dataset like CIFAR10 often assign high likelihoods to images from datasets with different objects (e.g., SVHN). We refine previous investigations of this failure at anomaly detection for invertible generative networks and provide a clear explanation of it as a combination of model bias and domain prior: Convolutional networks learn similar low-level feature distributions when trained on any natural image dataset and these low-level features dominate the likelihood. Hence, when the discriminative features between inliers and outliers are on a high-level, e.g., object shapes, anomaly detection becomes particularly challenging. To remove the negative impact of model bias and domain prior on detecting high-level differences, we propose two methods, first, using the log likelihood ratios of two identical models, one trained on the in-distribution data (e.g., CIFAR10) and the other one on a more general distribution of images (e.g., 80 Million Tiny Images). We also derive a novel outlier loss for the in-distribution network on samples from the more general distribution to further improve the performance. Secondly, using a multi-scale model like Glow, we show that low-level features are mainly captured at early scales. Therefore, using only the likelihood contribution of the final scale performs remarkably well for detecting high-level feature differences of the out-of-distribution and the in-distribution. This method is especially useful if one does not have access to a suitable general distribution. Overall, our methods achieve strong anomaly detection performance in the unsupervised setting, and only slightly underperform state-of-the-art classifier-based methods in the supervised setting. Code can be found at https://github.com/boschresearch/hierarchicalanomalydetection.

When it comes to the classification of brain signals in real-life applications, the training and the prediction data are often described by different distributions. Furthermore, diverse data sets, e.g., recorded from various subjects or tasks, can even exhibit distinct feature spaces. The fact that data that have to be classified are often only available in small amounts reinforces the need for techniques to generalize learned information, as performances of brain-computer interfaces (BCIs) are enhanced by increasing quantity of available data. In this paper, we apply transfer learning to a framework based on deep convolutional neural networks (deep ConvNets) to prove the transferability of learned patterns in error-related brain signals across different tasks. The experiments described in this paper demonstrate the usefulness of transfer learning, especially improving performances when only little data can be used to distinguish between erroneous and correct realization of a task. This effect could be delimited from a transfer of merely general brain signal characteristics, underlining the transfer of error-specific information. Furthermore, we could extract similar patterns in time-frequency analyses in identical channels, leading to selective high signal correlations between the two different paradigms. Classification on the intracranial data yields in median accuracies up to (81.50 ± 9.49)%. Decoding on only 10% of the data without pre-training reaches performances of (54.76 ± 3.56)%, compared to (64.95 ± 0.79)% with pre-training.

EEG is the most common signal source for noninvasive BCI applications. For such applications, the EEG signal needs to be decoded and translated into appropriate actions. A recently emerging EEG decoding approach is deep learning with Convolutional or Recurrent Neural Networks (CNNs, RNNs) with many different architectures already published. Here we present a novel framework for the large-scale evaluation of different deep-learning architectures on different EEG datasets. This framework comprises (i) a collection of EEG datasets currently including 100 examples (recording sessions) from six different classification problems, (ii) a collection of different EEG decoding algorithms, and (iii) a wrapper linking the decoders to the data as well as handling structured documentation of all settings and (hyper-) parameters and statistics, designed to ensure transparency and reproducibility. As an applications example we used our framework by comparing three publicly available CNN architectures: the Braindecode Deep4 ConvNet, Braindecode Shallow ConvNet, and two versions of EEGNet. We also show how our framework can be used to study similarities and differences in the performance of different decoding methods across tasks. We argue that the deep learning EEG framework as described here could help to tap the full potential of deep learning for BCI applications.

Deep learning techniques have revolutionized the field of machine learning and were recently successfully applied to various classification problems in noninvasive electroencephalography (EEG). However, these methods were so far only rarely evaluated for use in intracranial EEG. We employed convolutional neural networks (CNNs) to classify and characterize the error-related brain response as measured in 24 intracranial EEG recordings. Decoding accuracies of CNNs were significantly higher than those of a regularized linear discriminant analysis. Using time-resolved deep decoding, it was possible to classify errors in various regions in the human brain, and further to decode errors over 200 ms before the actual erroneous button press, e.g., in the precentral gyrus. Moreover, deeper networks performed better than shallower networks in distinguishing correct from error trials in all-channel decoding. In single recordings, up to 100 % decoding accuracy was achieved. Visualization of the networks' learned features indicated that multivariate decoding on an ensemble of channels yields related, albeit non-redundant information compared to single-channel decoding. In summary, here we show the usefulness of deep learning for both intracranial error decoding and mapping of the spatio-temporal structure of the human error processing network.