Consistency Models (CMs) have shown promise for efficient one-step generation. However, most existing CMs rely on manually designed discretization schemes, which can cause repeated adjustments for different noise schedules and datasets. To address this, we propose a unified framework for the automatic and adaptive discretization of CMs, formulating it as an optimization problem with respect to the discretization step. Concretely, during the consistency training process, we propose using local consistency as the optimization objective to ensure trainability by avoiding excessive discretization, and taking global consistency as a constraint to ensure stability by controlling the denoising error in the training target. We establish the trade-off between local and global consistency with a Lagrange multiplier. Building on this framework, we achieve adaptive discretization for CMs using the Gauss-Newton method. We refer to our approach as ADCMs. Experiments demonstrate that ADCMs significantly improve the training efficiency of CMs, achieving superior generative performance with minimal training overhead on both CIFAR-10 and ImageNet. Moreover, ADCMs exhibit strong adaptability to more advanced DM variants. Code is available at \url{https://github.com/rainstonee/ADCM}.

Finding a feasible solution is the primary concern in power system dispatch. This paper studies the feasibility condition of power system dispatch under a multistage robust optimization framework considering the non-anticipativity of dispatch policy, which is difficult to be expressed via explicit constraints. The multistage robust feasible regions (MRFRs) are defined as the sets in the state space containing all points that can maintain the robust feasibility in the next period against renewable and demand uncertainties; we give a polyhedral projection condition to characterize exact MRFRs in a recursive manner, which can be regarded as an analog of Bellman’s optimality condition. However, because the multistage dispatch problem of a bulk power system has a high-dimensional state space, the computation of exact MRFRs suffers from the curse of dimensionality. We propose an inner approximation method that identifies the maximal polyhedra that are embraced by the unknown exact MRFRs; we devise a computationally efficient algorithm to retrieve the hyperplane representation of the inner approximator. Finally, we discuss how MRFRs can be used in combination with existing approaches, such as dynamic programming and rolling horizon optimization. Numerical simulations on a modified IEEE 118-bus system verify the effectiveness and advantages of the proposed methodNote to Practitioners—This paper is motivated by the problem of maintaining the feasibility in power system multistage dispatch under renewable generation uncertainty. The proposed method regards the multistage dispatch as a sequential decision-making process and recursively defines the MRFR using polyhedral projection technique, which contains all the state points in each period that can guarantee the robust feasibility in the next period. To release the curses of dimensionality, the large-scale MRFR is approximated by an inner hyper-rectangle in the state space, which can be decomposed into independent intervals of state variables, such as the generation power range of coal-fired unit and the state-of-charge range of battery storage unit. These decoupled intervals are convenient for practical use and desired in the real-world power system. At the current stage, the calculation of MRFR requires a linear model; in the future research, integers will be addressed so that more facilities such as non-ideal energy storage units and fast-startup generators can be involved.

Monolithic large language models (LLMs) like GPT-4 have paved the way for modern generative AI applications. Training, serving, and maintaining monolithic LLMs at scale, however, remains prohibitively expensive and challenging. The disproportionate increase in compute-to-memory ratio of modern AI accelerators have created a memory wall, necessitating new methods to deploy AI. Recent research has shown that a composition of many smaller expert models, each with several orders of magnitude fewer parameters, can match or exceed the capabilities of monolithic LLMs. Composition of Experts (CoE) is a modular approach that lowers the cost and complexity of training and serving. However, this approach presents two key challenges when using conventional hardware: (1) without fused operations, smaller models have lower operational intensity, which makes high utilization more challenging to achieve; and (2) hosting a large number of models can be either prohibitively expensive or slow when dynamically switching between them. In this paper, we describe how combining CoE, streaming dataflow, and a three-tier memory system scales the AI memory wall. We describe Samba-CoE, a CoE system with 150 experts and a trillion total parameters. We deploy Samba-CoE on the SambaNova SN40L Reconfigurable Dataflow Unit (RDU) -a commercial dataflow accelerator architecture that has been codesigned for enterprise inference and training applications. The chip introduces a new three-tier memory system with on-chip distributed SRAM, on-package HBM, and off-package DDR DRAM. A dedicated inter-RDU network enables scaling up and out over multiple sockets. We demonstrate speedups ranging from 2× to 13× on various benchmarks running on eight RDU sockets compared with an unfused baseline. We show that for CoE inference deployments, the 8-socket RDU Node reduces machine footprint by up to 19 ×, speeds up model switching time by 15× to 31×, and achieves an overall speedup of 3.7× over a DGX H100 and 6.6× over a DGX A100.

The demand for indoor synthesis has increased significantly in recent years because of the emergence of computational design. This work designs and develops MyRoom, a Unity plugin that can import layout datasets for indoor synthesis and procedurally generate and interactively design indoor scenes. MyRoom enables users to easily edit and visualize non-intuitive layout description data, making it easier to generate and manipulate indoor scenes in Unity. MyRoom provides a user-friendly interface and powerful tools for designing and optimizing indoor layouts, including an automatic layout generator, making it ideal for game developers, interior designers, and researchers in the digital world-building industry. With MyRoom, users can streamline the process of creating high-quality indoor scenes in games and achieve their design goals efficiently.

Recent years, there has been growing interests in experience-driven procedural level generation. Various metrics have been formulated to model player experience and help generate personalised levels. In this work, we question whether experience metrics can adapt to agents with different personas. We start by reviewing existing metrics for evaluating game levels. Then, focusing on platformer games, we design a framework integrating various agents and evaluation metrics. Experimental studies on Super Mario Bros. indicate that using the same evaluation metrics but agents with different personas can generate levels for particular persona. It implies that, for simple games, using a game-playing agent of specific player archetype as a level tester is probably all we need to generate levels of diverse behaviour engagement.

Hyperspectral unmixing is an essential task in hyperspectral imagery applications. Because of the strong feature extract ability and satisfying performance, deep learning methods have been used for hyperspectral unmixing. However, there are still several problems in existing deep learning based spectral unmixing methods. Supervised learning methods can only accomplish a single task and lack a large amount of data for supervised learning. While the unsupervised learning unmixing methods are easily misled by the traditional way of initialization. In this paper, a weakly supervised deep convolutional neural network is proposed for hyperspectral unmixing. The experimental results show that competitive results can also be obtained by pretraining with a small number of samples, and weakly supervised learning still has potential for hyperspectral unmixing.

Hyperspectral unmixing is an essential task in hyperspectral imagery applications. Deep learning methods have been taken into hyperspectral unmixing because of its great feature extraction ability and better performance. However, there are several problems in existing deep learning based spectral unmixing methods. The networks are not deep enough to exploit their feature extraction capabilities in these unsupervised autoencoders based methods, and their effects are not stable. The main reason may be the limited prior information limited the ability of conducting the supervised method. In this manuscript, a semi-supervised deep learning based unmixing method is proposed. Unlike the existing methods, our model uses deeper neural networks without pooling layers, and the endmember spectrum are selected supervised from the original data, which uses nature and nurture cooperatively. The experimental results show that the proposed method achieves better performance and produces more accurate abundance maps, as well as higher quantitative results, compared with the current state-of-the-art deep learning unmixing algorithms.