Transformers have become the state-of-the-art neural network architecture across numerous domains of machine learning. This is partly due to their celebrated ability to transfer and to learn in-context based on few examples. Nevertheless, the mechanisms by which Transformers become in-context learners are not well understood and remain mostly an intuition. Here, we argue that training Transformers on auto-regressive tasks can be closely related to well-known gradient-based meta-learning formulations. We start by providing a simple weight construction that shows the equivalence of data transformations induced by 1) a single linear self-attention layer and by 2) gradient-descent (GD) on a regression loss. Motivated by that construction, we show empirically that when training self-attention-only Transformers on simple regression tasks either the models learned by GD and Transformers show great similarity or, remarkably, the weights found by optimization match the construction. Thus we show how trained Transformers implement gradient descent in their forward pass. This allows us, at least in the domain of regression problems, to mechanistically understand the inner workings of optimized Transformers that learn in-context. Furthermore, we identify how Transformers surpass plain gradient descent by an iterative curvature correction and learn linear models on deep data representations to solve non-linear regression tasks. Finally, we discuss intriguing parallels to a mechanism identified to be crucial for in-context learning termed induction-head (Olsson et al., 2022) and show how it could be understood as a specific case of in-context learning by gradient descent learning within Transformers.

从不同的随机初始化开始，经过随机梯度下降（SGD）训练的神经网络通常在功能上非常相似，从而提出了一个问题，即不同的SGD溶液之间是否存在有意义的差异。 Entezari等。最近猜想，尽管初始化不同，但在考虑到神经网络的置换不变性后，SGD发现的解决方案位于相同的损失谷中。具体而言，他们假设可以将SGD找到的任何两种解决方案排列，以使其参数之间的线性插值形成一条路径，而不会显着增加损失。在这里，我们使用一种简单但功能强大的算法来找到这样的排列，使我们能够获得直接的经验证据，证明该假设在完全连接的网络中是正确的。引人注目的是，我们发现在初始化时已经存在两个网络，并且平均它们随机，但适当排列的初始化的性能大大高于机会。相反，对于卷积架构，我们的证据表明该假设不存在。特别是在大型学习率制度中，SGD似乎发现了各种模式。

平衡系统是表达神经计算的有力方法。作为特殊情况，它们包括对神经科学和机器学习的最新兴趣模型，例如平衡复发性神经网络，深度平衡模型或元学习。在这里，我们提出了一个新的原则，用于学习具有时间和空间本地规则的此类系统。我们的原理将学习作为一个最不控制的问题，我们首先引入一个最佳控制器，以将系统带入解决方案状态，然后将学习定义为减少达到这种状态所需的控制量。我们表明，将学习信号纳入动力学作为最佳控制可以以先前未知的方式传输信用分配信息，避免将中间状态存储在内存中，并且不依赖无穷小的学习信号。在实践中，我们的原理可以使基于梯度的学习方法的强大绩效匹配，该方法应用于涉及复发性神经网络和元学习的一系列问题。我们的结果阐明了大脑如何学习并提供解决广泛的机器学习问题的新方法。

Machine learning models are typically evaluated by computing similarity with reference annotations and trained by maximizing similarity with such. Especially in the bio-medical domain, annotations are subjective and suffer from low inter- and intra-rater reliability. Since annotations only reflect the annotation entity's interpretation of the real world, this can lead to sub-optimal predictions even though the model achieves high similarity scores. Here, the theoretical concept of Peak Ground Truth (PGT) is introduced. PGT marks the point beyond which an increase in similarity with the reference annotation stops translating to better Real World Model Performance (RWMP). Additionally, a quantitative technique to approximate PGT by computing inter- and intra-rater reliability is proposed. Finally, three categories of PGT-aware strategies to evaluate and improve model performance are reviewed.

Cybercriminals are moving towards zero-day attacks affecting resource-constrained devices such as single-board computers (SBC). Assuming that perfect security is unrealistic, Moving Target Defense (MTD) is a promising approach to mitigate attacks by dynamically altering target attack surfaces. Still, selecting suitable MTD techniques for zero-day attacks is an open challenge. Reinforcement Learning (RL) could be an effective approach to optimize the MTD selection through trial and error, but the literature fails when i) evaluating the performance of RL and MTD solutions in real-world scenarios, ii) studying whether behavioral fingerprinting is suitable for representing SBC's states, and iii) calculating the consumption of resources in SBC. To improve these limitations, the work at hand proposes an online RL-based framework to learn the correct MTD mechanisms mitigating heterogeneous zero-day attacks in SBC. The framework considers behavioral fingerprinting to represent SBCs' states and RL to learn MTD techniques that mitigate each malicious state. It has been deployed on a real IoT crowdsensing scenario with a Raspberry Pi acting as a spectrum sensor. More in detail, the Raspberry Pi has been infected with different samples of command and control malware, rootkits, and ransomware to later select between four existing MTD techniques. A set of experiments demonstrated the suitability of the framework to learn proper MTD techniques mitigating all attacks (except a harmfulness rootkit) while consuming <1 MB of storage and utilizing <55% CPU and <80% RAM.

Mixtures of von Mises-Fisher distributions can be used to cluster data on the unit hypersphere. This is particularly adapted for high-dimensional directional data such as texts. We propose in this article to estimate a von Mises mixture using a l 1 penalized likelihood. This leads to sparse prototypes that improve clustering interpretability. We introduce an expectation-maximisation (EM) algorithm for this estimation and explore the trade-off between the sparsity term and the likelihood one with a path following algorithm. The model's behaviour is studied on simulated data and, we show the advantages of the approach on real data benchmark. We also introduce a new data set on financial reports and exhibit the benefits of our method for exploratory analysis.

The future of population-based breast cancer screening is likely personalized strategies based on clinically relevant risk models. Mammography-based risk models should remain robust to domain shifts caused by different populations and mammographic devices. Modern risk models do not ensure adaptation across vendor-domains and are often conflated to unintentionally rely on both precursors of cancer and systemic/global mammographic information associated with short- and long-term risk, respectively, which might limit performance. We developed a robust, cross-vendor model for long-term risk assessment. An augmentation-based domain adaption technique, based on flavorization of mammographic views, ensured generalization to an unseen vendor-domain. We trained on samples without diagnosed/potential malignant findings to learn systemic/global breast tissue features, called mammographic texture, indicative of future breast cancer. However, training so may cause erratic convergence. By excluding noise-inducing samples and designing a case-control dataset, a robust ensemble texture model was trained. This model was validated in two independent datasets. In 66,607 Danish women with flavorized Siemens views, the AUC was 0.71 and 0.65 for prediction of interval cancers within two years (ICs) and from two years after screening (LTCs), respectively. In a combination with established risk factors, the model's AUC increased to 0.68 for LTCs. In 25,706 Dutch women with Hologic-processed views, the AUCs were not different from the AUCs in Danish women with flavorized views. The results suggested that the model robustly estimated long-term risk while adapting to an unseen processed vendor-domain. The model identified 8.1% of Danish women accounting for 20.9% of ICs and 14.2% of LTCs.

Quaternion valued neural networks experienced rising popularity and interest from researchers in the last years, whereby the derivatives with respect to quaternions needed for optimization are calculated as the sum of the partial derivatives with respect to the real and imaginary parts. However, we can show that product- and chain-rule does not hold with this approach. We solve this by employing the GHRCalculus and derive quaternion backpropagation based on this. Furthermore, we experimentally prove the functionality of the derived quaternion backpropagation.

The goal of this paper is to detect objects by exploiting their interrelationships. Rather than relying on predefined and labeled graph structures, we infer a graph prior from object co-occurrence statistics. The key idea of our paper is to model object relations as a function of initial class predictions and co-occurrence priors to generate a graph representation of an image for improved classification and bounding box regression. We additionally learn the object-relation joint distribution via energy based modeling. Sampling from this distribution generates a refined graph representation of the image which in turn produces improved detection performance. Experiments on the Visual Genome and MS-COCO datasets demonstrate our method is detector agnostic, end-to-end trainable, and especially beneficial for rare object classes. What is more, we establish a consistent improvement over object detectors like DETR and Faster-RCNN, as well as state-of-the-art methods modeling object interrelationships.

Neuromorphic systems require user-friendly software to support the design and optimization of experiments. In this work, we address this need by presenting our development of a machine learning-based modeling framework for the BrainScaleS-2 neuromorphic system. This work represents an improvement over previous efforts, which either focused on the matrix-multiplication mode of BrainScaleS-2 or lacked full automation. Our framework, called hxtorch.snn, enables the hardware-in-the-loop training of spiking neural networks within PyTorch, including support for auto differentiation in a fully-automated hardware experiment workflow. In addition, hxtorch.snn facilitates seamless transitions between emulating on hardware and simulating in software. We demonstrate the capabilities of hxtorch.snn on a classification task using the Yin-Yang dataset employing a gradient-based approach with surrogate gradients and densely sampled membrane observations from the BrainScaleS-2 hardware system.