Recent advancements in generative models have highlighted the crucial role of image tokenization in the efficient synthesis of high-resolution images. Tokenization, which transforms images into latent representations, reduces computational demands compared to directly processing pixels and enhances the effectiveness and efficiency of the generation process. Prior methods, such as VQGAN, typically utilize 2D latent grids with fixed downsampling factors. However, these 2D tokenizations face challenges in managing the inherent redundancies present in images, where adjacent regions frequently display similarities. To overcome this issue, we introduce Transformer-based 1-Dimensional Tokenizer (TiTok), an innovative approach that tokenizes images into 1D latent sequences. TiTok provides a more compact latent representation, yielding substantially more efficient and effective representations than conventional techniques. For example, a 256 256 3 image can be reduced to just 32 discrete tokens, a significant reduction from the 256 or 1024 tokens obtained by prior methods. Despite its compact nature, TiTok achieves competitive performance to state-of-the-art approaches.

Recent controllable generation approaches such as FreeControl [24] and Diffusion Self-Guidance [7] bring fine-grained spatial and appearance control to text-toimage (T2I) diffusion models without training auxiliary modules.

In offline reinforcement learning (RL), a learner leverages prior logged data to learn a good policy without interacting with the environment. A major challenge in applying such methods in practice is the lack of both theoretically principled and practical tools for model selection and evaluation. To address this, we study the problem of model selection in offline RL with value function approximation. The learner is given a nested sequence of model classes to minimize squared Bellman error and must select among these to achieve a balance between approximation and estimation error of the classes. We propose the first model selection algorithm for offline RL that achieves minimax rate-optimal oracle inequalities up to logarithmic factors.

The task of layout-to-image generation involves synthesizing images based on the captions of objects and their spatial positions. Existing methods still struggle in complex layout generation, where common bad cases include object missing, inconsistent lighting, conflicting view angles, etc. To effectively address these issues, we propose a Hierarchical Controllable (HiCo) diffusion model for layout-to-image generation, featuring object seperable conditioning branch structure. Our key insight is to achieve spatial disentanglement through hierarchical modeling of layouts. We use a multi branch structure to represent hierarchy and aggregate them in fusion module.

This paper studies quasi-Newton methods for solving nonlinear equations. We propose block variants of both good and bad Broyden's methods, which enjoy explicit local superlinear convergence rates. Our block good Broyden's method has a faster condition-number-free convergence rate than existing Broyden's methods because it takes the advantage of multiple rank modification on Jacobian estimator. On the other hand, our block bad Broyden's method directly estimates the inverse of the Jacobian provably, which reduces the computational cost of the iteration. Our theoretical results provide some new insights on why good Broyden's method outperforms bad Broyden's method in most of the cases. The empirical results also demonstrate the superiority of our methods and validate our theoretical analysis.

Detecting dependencies among variables is a fundamental task across scientific disciplines. We propose a novel neural network model for graph structure inference, which aims to learn a mapping from observational data to the corresponding underlying dependence structures. The model is trained with variably shaped and coupled simulated input data and requires only a single forward pass through the trained network for inference. Central to our approach is a novel bilinear attention mechanism (BAM) operating on covariance matrices of transformed data while respecting the geometry of the manifold of symmetric positive definite (SPD) matrices. Inspired by graphical lasso methods, our model optimizes over continuous graph representations in the SPD space, where inverse covariance matrices encode conditional independence relations. Empirical evaluations demonstrate the robustness of our method in detecting diverse dependencies, excelling in undirected graph estimation and showing competitive performance in completed partially directed acyclic graph estimation via a novel two-step approach. The trained model effectively detects causal relationships and generalizes well across different functional forms of nonlinear dependencies.

Spiking neural networks (SNNs), inspired by biological processes, use spike signals for inter-layer communication, presenting an energy-efficient alternative to traditional neural networks. To realize the theoretical advantages of SNNs in energy efficiency, it is essential to deploy them onto neuromorphic chips. On clock-driven synchronous chips, employing shorter time steps can enhance energy efficiency but reduce SNN performance. Compared to the clock-driven synchronous chip, the event-driven asynchronous chip achieves much lower energy consumption but only supports some specific network operations. Recently, a series of SNN projects have achieved tremendous success, significantly improving the SNN's performance. However, event-driven asynchronous chips do not support some of the proposed structures, making it impossible to integrate these SNNs into asynchronous hardware. In response to these problems, we propose the Spiking Token Mixer (STMixer) architecture, which consists exclusively of operations supported by asynchronous scenarios, including convolutional, fully connected layers and residual paths. Our series of experiments also demonstrates that STMixer achieves performance on par with spiking transformers in synchronous scenarios with very low timesteps. This indicates its ability to achieve the same level of performance with lower power consumption in synchronous scenarios.

Organization In this supplementary file, we provide in-depth descriptions of the materials that are not covered in the main paper, and report additional experimental results. The document is organized as follows: Section A-Related work. Neural Architecture Search (NAS) was introduced to ease the process of manually designing complex neural networks. Early NAS [1] efforts employed a brute force approach by training candidate architectures and using their accuracy as a proxy for discovering superior designs. One-shot NAS methods [5, 6, 7] further reduced the cost by training large supernetworks and identifying high-accuracy subnetworks, often generated from pre-trained models. Nevertheless, as search spaces expand with architectural innovations [8, 9], more efficient methods are necessary to predict neural network accuracy in vast design spaces. Recent mathematical programming (MP) based NAS methods [10, 11] are noteworthy, as they transform multi-objective NAS problems into mathematical programming solutions.