This plot shows a TSNE of CLIP's pooled output for the same

Cross Attention is a popular method for retrieving information from a set of context tokens for making predictions. At inference time, for each prediction, Cross Attention scans the full set of $\mathcal{O}(N)$ tokens. In practice, however, often only a small subset of tokens are required for good performance. Methods such as Perceiver IO are cheap at inference as they distill the information to a smaller-sized set of latent tokens $L < N$ on which cross attention is then applied, resulting in only $\mathcal{O}(L)$ complexity. However, in practice, as the number of input tokens and the amount of information to distill increases, the number of latent tokens needed also increases significantly. In this work, we propose Tree Cross Attention (TCA) - a module based on Cross Attention that only retrieves information from a logarithmic $\mathcal{O}(\log(N))$ number of tokens for performing inference. TCA organizes the data in a tree structure and performs a tree search at inference time to retrieve the relevant tokens for prediction. Leveraging TCA, we introduce ReTreever, a flexible architecture for token-efficient inference. We show empirically that Tree Cross Attention (TCA) performs comparable to Cross Attention across various classification and uncertainty regression tasks while being significantly more token-efficient. Furthermore, we compare ReTreever against Perceiver IO, showing significant gains while using the same number of tokens for inference.

Deep learning surrogate models are being increasingly used in accelerating scientific simulations as a replacement for costly conventional numerical techniques. However, their use remains a significant challenge when dealing with real-world complex examples. In this work, we demonstrate three types of neural network architectures for efficient learning of highly non-linear deformations of solid bodies. The first two architectures are based on the recently proposed CNN U-NET and MAgNET (graph U-NET) frameworks which have shown promising performance for learning on mesh-based data. The third architecture is Perceiver IO, a very recent architecture that belongs to the family of attention-based neural networks--a class that has revolutionised diverse engineering fields and is still unexplored in computational mechanics. We study and compare the performance of all three networks on two benchmark examples, and show their capabilities to accurately predict the non-linear mechanical responses of soft bodies.

Gene regulation is a dynamic process that connects genotype and phenotype. Given the difficulty of physically mapping mammalian gene circuitry, we require new computational methods to learn regulatory rules. Natural language is a valuable analogy to the communication of regulatory control. Machine learning systems model natural language by explicitly learning context dependencies between words. We propose a similar system applied to single-cell RNA expression profiles to learn context dependencies between genes. Our model, Exceiver, is trained across a diversity of cell types using a self-supervised task formulated for discrete count data, accounting for feature sparsity. We found agreement between the similarity profiles of latent sample representations and learned gene embeddings with respect to biological annotations. We evaluated Exceiver on a new dataset and a downstream prediction task and found that pretraining supports transfer learning. Our work provides a framework to model gene regulation on a single-cell level and transfer knowledge to downstream tasks.

Recent work has seen the development of general purpose neural architectures that can be trained to perform tasks across diverse data modalities. General purpose models typically make few assumptions about the underlying data-structure and are known to perform well in the large-data regime. At the same time, there has been growing interest in modular neural architectures that represent the data using sparsely interacting modules. These models can be more robust out-of-distribution, computationally efficient, and capable of sample-efficient adaptation to new data. However, they tend to make domain-specific assumptions about the data, and present challenges in how module behavior (i.e., parameterization) and connectivity (i.e., their layout) can be jointly learned. In this work, we introduce a general purpose, yet modular neural architecture called Neural Attentive Circuits (NACs) that jointly learns the parameterization and a sparse connectivity of neural modules without using domain knowledge. NACs are best understood as the combination of two systems that are jointly trained end-to-end: one that determines the module configuration and the other that executes it on an input. We demonstrate qualitatively that NACs learn diverse and meaningful module configurations on the NLVR2 dataset without additional supervision. Quantitatively, we show that by incorporating modularity in this way, NACs improve upon a strong non-modular baseline in terms of low-shot adaptation on CIFAR and CUBs dataset by about 10%, and OOD robustness on Tiny ImageNet-R by about 2.5%. Further, we find that NACs can achieve an 8x speedup at inference time while losing less than 3% performance. Finally, we find NACs to yield competitive results on diverse data modalities spanning point-cloud classification, symbolic processing and text-classification from ASCII bytes, thereby confirming its general purpose nature.

Multimodal machine learning has been widely studied for the development of general intelligence. Recently, the remarkable multimodal algorithms, the Perceiver and Perceiver IO, show competitive results for diverse dataset domains and tasks. However, recent works, Perceiver and Perceiver IO, have focused on heterogeneous modalities, including image, text, and speech, and there are few research works for graph structured datasets. A graph is one of the most generalized dataset structures, and we can represent the other dataset, including images, text, and speech, as graph structured data. A graph has an adjacency matrix different from other dataset domains such as text and image, and it is not trivial to handle the topological information, relational information, and canonical positional information. In this study, we provide a Graph Perceiver IO, the Perceiver IO for the graph structured dataset. We keep the main structure of the Graph Perceiver IO as the Perceiver IO because the Perceiver IO already handles the diverse dataset well, except for the graph structured dataset. The Graph Perceiver IO is a general method, and it can handle diverse datasets such as graph structured data as well as text and images. Comparing the graph neural networks, the Graph Perceiver IO requires a lower complexity, and it can incorporate the local and global information efficiently. We show that Graph Perceiver IO shows competitive results for diverse graph-related tasks, including node classification, graph classification, and link prediction.

Recently, DeepMind has open-sourced Perceiver IO–a general-purpose deep learning model architecture that can handle many different types of inputs and outputs. This "drop-in" replacement for Transformers is powerful enough to outperform baseline models without being constrained by domain knowledge. A new preprint on arXiv describes Perceiver IO, a more general version of the AI architecture that can produce many different outputs from multiple inputs. This means it is applicable to real-world domains like language and vision as well as difficult games like StarCraft II. Unlike Perceiver, Perceiver IO is an advanced model that overcomes the limitation of only being able to produce very simple outputs by learning how to flexibly query the latent space.