This paper describes Asterisk, a compact GPT-based model for generating text embeddings. The model uses a minimalist architecture with two layers, two attention heads, and 256 embedding dimensions. By applying knowledge distillation from larger pretrained models, we explore the trade-offs between model size and performance while minimizing computational and memory requirements. The model is primarily evaluated and optimized for classification tasks, with experimental results showing its moderate performance in zero-shot classification across various downstream applications. With additional configuration, the model performance can approach or even surpass that of larger architectures on specific classification tasks.

Continuous-output neural machine translation (CoNMT) replaces the discrete next-word prediction problem with an embedding prediction. The semantic structure of the target embedding space (i.e., closeness of related words) is intuitively believed to be crucial. We challenge this assumption and show that completely random output embeddings can outperform laboriously pretrained ones, especially on larger datasets. Further investigation shows this surprising effect is strongest for rare words, due to the geometry of their embeddings. We shed further light on this finding by designing a mixed strategy that combines random and pre-trained embeddings for different tokens.

Contrastive learning is a self-supervised representation learning framework, where two positive views generated through data augmentation are made similar by an attraction force in a data representation space, while a repulsive force makes them far from negative examples. Non-contrastive learning, represented by BYOL and SimSiam, further gets rid of negative examples and improves computational efficiency. While learned representations may collapse into a single point due to the lack of the repulsive force at first sight, Tian et al. (2021) revealed through the learning dynamics analysis that the representations can avoid collapse if data augmentation is sufficiently stronger than regularization. However, their analysis does not take into account commonly-used feature normalization, a normalizer before measuring the similarity of representations, and hence excessively strong regularization may collapse the dynamics, which is an unnatural behavior under the presence of feature normalization. Therefore, we extend the previous theory based on the L2 loss by considering the cosine loss, which involves feature normalization. We show that the cosine loss induces sixth-order dynamics (while the L2 loss induces a third-order one), in which a stable equilibrium dynamically emerges even if there are only collapsed solutions with given initial parameters. Thus, we offer a new understanding that feature normalization plays an important role in robustly preventing the dynamics collapse.

Group-equivariant neural networks have emerged as a data-efficient approach to solve classification and regression tasks, while respecting the relevant symmetries of the data. However, little work has been done to extend this paradigm to the unsupervised and generative domains. Here, we present Holographic-(Variational) Auto Encoder (H-(V)AE), a fully end-to-end SO(3)-equivariant (variational) autoencoder in Fourier space, suitable for unsupervised learning and generation of data distributed around a specified origin in 3D. H-(V)AE is trained to reconstruct the spherical Fourier encoding of data, learning in the process a low-dimensional representation of the data (i.e., a latent space) with a maximally informative rotationally invariant embedding alongside an equivariant frame describing the orientation of the data. We extensively test the performance of H-(V)AE on diverse datasets. We show that the learned latent space efficiently encodes the categorical features of spherical images. Moreover, H-(V)AE's latent space can be used to extract compact embeddings for protein structure microenvironments, and when paired with a Random Forest Regressor, it enables state-of-the-art predictions of protein-ligand binding affinity.

Segmentation of COVID-19 lesions can assist physicians in better diagnosis and treatment of COVID-19. However, there are few relevant studies due to the lack of detailed information and high-quality annotation in the COVID-19 dataset. To solve the above problem, we propose C2FVL, a Coarse-to-Fine segmentation framework via Vision-Language alignment to merge text information containing the number of lesions and specific locations of image information. The introduction of text information allows the network to achieve better prediction results on challenging datasets. We conduct extensive experiments on two COVID-19 datasets including chest X-ray and CT, and the results demonstrate that our proposed method outperforms other state-of-the-art segmentation methods.

It may sound a bit obvious, but sometimes we miss the power of this kind of solution to solve small data problems -- I'm talking about Data Augmentation. The idea behind Data Augmentation is -- Points nearby a certain point (in hyperspace) represent a similar behavior. For eg: An image of a dog with increased contrast or brightness is still an image of a dog. Let's talk about SMOTE in more detail. To simplify we can define the concept behind SMOTE as "Birds of the same feather flock together", which reduced in data terms means that data-points that are close to each other in the hyperspace represent similar behavior so they can be approximated as new data-points for the dataset.

In this work, we perform a wide variety of experiments with different Deep Learning architectures in small data conditions. We show that model complexity is a critical factor when only a few samples per class are available. Differently from the literature, we improve the state of the art using low complexity models. We show that standard convolutional neural networks with relatively few parameters are effective in this scenario. In many of our experiments, low complexity models outperform state-of-the-art architectures. Moreover, we propose a novel network that uses an unsupervised loss to regularize its training. Such architecture either improves the results either performs comparably well to low capacity networks. Surprisingly, experiments show that the dynamic data augmentation pipeline is not beneficial in this particular domain. Statically augmenting the dataset might be a promising research direction while dropout maintains its role as a good regularizer.

It promises to solve your most complicated problems for the small price of an enormous amount of data. The only problem is you are not working at Google nor Facebook and data are scarce. So what are you to do? Can you still leverage the power of deep learning or are you out of luck? Let's take a look at how you might be able to leverage deep learning even with limited data and why I think this might be one of the most exciting areas of future research. Before we discuss methods for leveraging deep learning for your limited data, please step back from the neural networks and build a simple baseline.

Two things seem to be indisputable in the contemporary deep learning discourse: 1. The categorical cross-entropy loss after softmax activation is the method of choice for classification. 2. Training a CNN classifier from scratch on small datasets does not work well. In contrast to this, we show that the cosine loss function provides significantly better performance than cross-entropy on datasets with only a handful of samples per class. For example, the accuracy achieved on the CUB-200-2011 dataset without pre-training is by 30% higher than with the cross-entropy loss. Further experiments on four other popular datasets confirm our findings. Moreover, we show that the classification performance can be improved further by integrating prior knowledge in the form of class hierarchies, which is straightforward with the cosine loss.