Autoregressive models, such as the GPT family, use a fixed order, usually left-to-right, to generate sequences. However, this is not a necessity. In this paper, we challenge this assumption and show that by simply adding a positional encoding for the output, this order can be modulated on-the-fly per-sample which offers key advantageous properties. It allows for the sampling of and conditioning on arbitrary subsets of tokens, and it also allows sampling in one shot multiple tokens dynamically according to a rejection strategy, leading to a sub-linear number of model evaluations. We evaluate our method across various domains, including language modeling, path-solving, and aircraft vertical rate prediction, decreasing the number of steps required for generation by an order of magnitude.

Transformer-based architectures are the model of choice for natural language understanding, but they come at a significant cost, as they have quadratic complexity in the input length, require a lot of training data, and can be difficult to tune. In the pursuit of lower costs, we investigate simple MLP-based architectures. We find that existing architectures such as MLPMixer, which achieves token mixing through a static MLP applied to each feature independently, are too detached from the inductive biases required for natural language understanding. In this paper, we propose a simple variant, HyperMixer, which forms the token mixing MLP dynamically using hypernetworks. Empirically, we demonstrate that our model performs better than alternative MLP-based models, and on par with Transformers. In contrast to Transformers, HyperMixer achieves these results at substantially lower costs in terms of processing time, training data, and hyperparameter tuning.

We propose attention-based modeling of quantities at arbitrary spatial points conditioned on related measurements at different locations. Our approach adapts a transformer-encoder to process measurements and read-out positions together. Attention-based models exhibit excellent performance across domains, which makes them an interesting candidate for modeling data irregularly sampled in space. We introduce a novel encoding strategy that applies the same transformation to the measurements and read-out positions, after which they are combined with encoded measurement values instead of relying on two different mappings. Efficiently learning input-output mappings from irregularly-spaced data is a fundamental challenge in modeling physical phenomena. To evaluate the effectiveness of our model, we conduct experiments on diverse problem domains, including high-altitude wind nowcasting, two-days weather forecasting, fluid dynamics, and heat diffusion. Our attention-based model consistently outperforms state-of-the-art models, such as Graph Element Networks and Conditional Neural Processes, for modeling irregularly sampled data. Notably, our model reduces root mean square error (RMSE) for wind nowcasting, improving from 9.24 to 7.98 and for a heat diffusion task from .126 to .084. We hypothesize that this superior performance can be attributed to the enhanced flexibility of our latent representation and the improved data encoding technique. To support our hypothesis, we design a synthetic experiment that reveals excessive bottlenecking in the latent representations of alternative models, which hinders information utilization and impedes training.