However, existing contrastive learning methods primarily focus on one single data level, which fails to fully exploit the intricate nature of medical time series.

These authors contributed equally to this work.

By pairing historical and future trajectories and applying contrastive learning on the encoded feature space, we enforce same-space consistency constraints.

Hackers made death threats against this security researcher. In April 2024, a mysterious someone using the online handles "Waifu" and "Judische" began posting death threats on Telegram and Discord channels aimed at a cybersecurity researcher named Allison Nixon. These anonymous personas targeted Nixon because she had become a formidable threat: As chief research officer at the cyber investigations firm Unit 221B, named after Sherlock Holmes's apartment, she had built a career tracking cybercriminals and helping get them arrested. Though she'd done this work for more than a decade, Nixon couldn't understand why the person behind the accounts was suddenly threatening her. And although she had taken an interest in the Waifu persona in years past for crimes he boasted about committing, he hadn't been on her radar for a while when the threats began, because she was tracking other targets. Now Nixon resolved to unmask Waifu/Judische and others responsible for the death threats--and take them down for crimes they admitted to committing.

In the rapidly evolving field of deep learning, the demand for models that are both expressive and computationally efficient has never been more critical. This paper introduces Orchid, a novel architecture designed to address the quadratic complexity of traditional attention mechanisms without compromising the ability to capture long-range dependencies and in-context learning. At the core of this architecture lies a new data-dependent global convolution layer, which contextually adapts its kernel conditioned on input sequence using a dedicated conditioning neural network. We design two simple conditioning networks that maintain shift equivariance in our data-dependent convolution operation. The dynamic nature of the proposed convolution kernel grants Orchid high expressivity while maintaining quasilinear scalability for long sequences. We evaluate the proposed model across multiple domains, including language modeling and image classification, to highlight its performance and generality. Our experiments demonstrate that this architecture not only outperforms traditional attention-based architectures such as BERT and Vision Transformers with smaller model sizes, but also extends the feasible sequence length beyond the limitations of the dense attention layers. This achievement represents a significant step towards more efficient and scalable deep learning models for sequence modeling.