Probabilistic Circuits (PCs) models are emerging popular tractable probabilistic models. Their internal connections are represented in the form of directed acyclic graphs (DAGs) with sum nodes and product nodes, ensuring their internal parameter efficiency and model expressiveness in terms of probabilistic inference. Despite these algorithmic advantages, executing PC still faces graph structure deployment issues. PyJuice on GPU with the block-sparse parallel computation methods causes a parallelism-sparsity gap, while DAG-style processing does not take advantage of the repetitive characteristics of PC internal nodes, resulting in low throughput. To address this challenge, this work proposes the ESTroM, an efficient architecture that provides novel graph-element (nodes/edges) parallelism with sparsity-aware compilation. Through analysis of the sum/product node computational requirements, ESTroM core uses compressed matrices for sum/product nodes DAG representations, edge-based dataflow for product node processing, and node-based dataflow for sum node processing. With intra-core rewind and intercore multicast optimizations, we develop a prototype ESTrom chip and a demonstrative system for a PC-based neural lossless compression application. Our ablation experiments show ESTrom offers a speed improvement of$2.11 \sim 3.79 \times$compared to the state-of-the-art DAG processing unit (DPU)-v2 with the same computing resources. Under various typical PC structures, ESTrom achieves a speedup of$18.7 \times$compared to DPU-v2 and$3.9 \times$compared to NVIDIA RTX 4090 GPU with PyJuice framework. In terms of neural lossless compression, ESTroM demonstrates a$1.39 \times$improvement in compression ratio compared to the industrial-standard Z-standard (Zstd) algorithms with the highest compression level, while offering$16.3 \sim 65.2 \times$improvement in compression speed compared to Zstd on Intel Xeon Gold 6230. In a nutshell, this work develops novel graph element parallelism and element-flow architecture theory with practical prototype chips and systems, revealing a new hardware-perspective path for the “scaling law” of emerging tractable probabilistic models.

Mobile edge computing (MEC) is a promising technique to improve the computational capacity of smart devices (SDs) in Internet of Things (IoT). However, the performance of MEC is restricted due to its fixed location and limited service scope. Hence, we investigate an unmanned aerial vehicle (UAV)assisted MEC system, where multiple UAVs are dispatched and each UAV can simultaneously provide computing service for multiple SDs. To improve the performance of system, we formulated a UAV-based trajectory control and resource allocation multi-objective optimization problem (TCRAMOP) to simultaneously maximize the offloading number of UAVs and minimize total offloading delay and total energy consumption of UAVs by optimizing the flight paths of UAVs as well as the computing resource allocated to served SDs. Then, consider that the solution of TCRAMOP requires continuous decision-making and the system is dynamic, we propose an enhanced deep reinforcement learning (DRL) algorithm, namely, distributed proximal policy optimization with imitation learning (DPPOIL). This algorithm incorporates the generative adversarial imitation learning technique to improve the policy performance. Simulation results demonstrate the effectiveness of our proposed DPPOIL and prove that the learned strategy of DPPOIL is better compared with other baseline methods.

Non-autoregressive (NAR) generative models are valuable because they can handle diverse conditional generation tasks in a more principled way than their autoregressive (AR) counterparts, which are constrained by sequential dependency requirements. Recent advancements in NAR models, such as diffusion language models, have demonstrated superior performance in unconditional generation compared to AR models (e.g., GPTs) of similar sizes. However, such improvements do not always lead to improved conditional generation performance. We show that a key reason for this gap is the difficulty in generalizing to conditional probability queries unseen during training. As a result, strong unconditional generation performance does not guarantee high-quality conditional generation. This paper proposes Tractable Transformers (Tracformer), a Transformer-based generative model that is more robust to different conditional generation tasks. Unlike existing models that rely solely on global contextual features derived from full inputs, Tracformers incorporate a sparse Transformer encoder to capture both local and global contextual information. This information is routed through a decoder for conditional generation. Empirical results demonstrate that Tracformers achieve state-of-the-art conditional generation performance on text modeling compared to recent diffusion and AR model baselines.

Masked generative models (MGMs) have emerged as a powerful framework for image synthesis, combining parallel decoding with strong bidirectional context modeling. However, generating high-quality samples typically requires many iterative decoding steps, resulting in high inference costs. A straightforward way to speed up generation is by decoding more tokens in each step, thereby reducing the total number of steps. However, when many tokens are decoded simultaneously, the model can only estimate the univariate marginal distributions independently, failing to capture the dependency among them. As a result, reducing the number of steps significantly compromises generation fidelity. In this work, we introduce ReCAP (Reused Context-Aware Prediction), a plug-and-play module that accelerates inference in MGMs by constructing low-cost steps via reusing feature embeddings from previously decoded context tokens. ReCAP interleaves standard full evaluations with lightweight steps that cache and reuse context features, substantially reducing computation while preserving the benefits of fine-grained, iterative generation. We demonstrate its effectiveness on top of three representative MGMs (MaskGIT, MAGE, and MAR), including both discrete and continuous token spaces and covering diverse architectural designs. In particular, on ImageNet256 class-conditional generation, ReCAP achieves up to 2.4$\times$ faster inference than the base model with minimal performance drop, and consistently delivers better efficiency–fidelity trade-offs under various generation settings.

A popular paradigm for offline Reinforcement Learning (RL) tasks is to first fit the offline trajectories to a sequence model, and then prompt the model for actions that lead to high expected return. In addition to obtaining accurate sequence models, this paper highlights that tractability, the ability to exactly and efficiently answer various probabilistic queries, plays an important role in offline RL. Specifically, due to the fundamental stochasticity from the offline data-collection policies and the environment dynamics, highly non-trivial conditional/constrained generation is required to elicit rewarding actions. While it is still possible to approximate such queries, we observe that such crude estimates undermine the benefits brought by expressive sequence models. To overcome this problem, this paper proposes Trifle (Tractable Inference for Offline RL), which leverages modern tractable generative models to bridge the gap between good sequence models and high expected returns at evaluation time. Empirically, Trifle achieves $7$ state-of-the-art scores and the highest average scores in $9$ Gym-MuJoCo benchmarks against strong baselines. Further, Trifle significantly outperforms prior approaches in stochastic environments and safe RL tasks with minimum algorithmic modifications.

This paper presents OmniJARVIS, a novel Vision-Language-Action (VLA) model for open-world instruction-following agents in Minecraft. Compared to prior works that either emit textual goals to separate controllers or produce the control command directly, OmniJARVIS seeks a different path to ensure both strong reasoning and efficient decision-making capabilities via unified tokenization of multimodal interaction data. First, we introduce a self-supervised approach to learn a behavior encoder that produces discretized tokens for behavior trajectories $\tau = \{o_0, a_0, \dots\}$ and an imitation learning policy decoder conditioned on these tokens. These additional behavior tokens will be augmented to the vocabulary of pretrained Multimodal Language Models. With this encoder, we then pack long-term multimodal interactions involving task instructions, memories, thoughts, observations, textual responses, behavior trajectories, etc into unified token sequences and model them with autoregressive transformers. Thanks to the semantically meaningful behavior tokens, the resulting VLA model, OmniJARVIS, can reason (by producing chain-of-thoughts), plan, answer questions, and act (by producing behavior tokens for the imitation learning policy decoder). OmniJARVIS demonstrates excellent performances on a comprehensive collection of atomic, programmatic, and open-ended tasks in open-world Minecraft. Our analysis further unveils the crucial design principles in interaction data formation, unified tokenization, and its scaling potentials. The dataset, models, and code will be released at https://craftjarvis.org/OmniJARVIS.

The point of interest (POI) recommendation algorithm in location based social network (LBSN) can assist people to find more appealing locations and satisfy their specific demands. However, it is challengeable to infer user’s preference due to the sparsity of the user’s check-in data. To address the problem and improve recommendation performance, this paper proposes an improved context-aware weighted matrix factorization algorithm for POI recommendation (ICWMF). It takes advantage of time factor, geographical information, and social relationship to obtain user’s preference for locations. Firstly, the Ebbinghaus forgetting curve is employed to model the influence of time attenuation, so as to reflect that user preferences change over time. In order to assign dynamic weights to unvisited POI and infer user preference, we build the implicit feedback term by modeling the geographical influence from user perspective and the social relationship. In addition, the Gaussian model is employed to construct proximity location relationship to represent the probability of locations being discovered by users. Then, it is taken as the regularization term to avoid overfitting. Finally, the objective function of weighted matrix factorization is reconstructed with the implicit feedback term and the regularization term we designed. ICWMF naturally learns two potential feature matrices during weighted matrix decomposition based on new designed objective function to achieve better recommendation results. The results of simulation experiments on Brightkite and Gowalla dataset indicate that ICWMF outperforms other four comparison methods in terms of precision and recall.

Probabilistic Circuits (PCs) are a general and unified computational framework for tractable probabilistic models that support efficient computation of various inference tasks (e.g., computing marginal probabilities). Towards enabling such reasoning capabilities in complex real-world tasks, Liu et al. (2022) propose to distill knowledge (through latent variable assignments) from less tractable but more expressive deep generative models. However, it is still unclear what factors make this distillation work well. In this paper, we theoretically and empirically discover that the performance of a PC can exceed that of its teacher model. Therefore, instead of performing distillation from the most expressive deep generative model, we study what properties the teacher model and the PC should have in order to achieve good distillation performance. This leads to a generic algorithmic improvement as well as other data-type-specific ones over the existing latent variable distillation pipeline. Empirically, we outperform SoTA TPMs by a large margin on challenging image modeling benchmarks. In particular, on ImageNet32, PCs achieve 4.06 bits-per-dimension, which is only 0.34 behind variational diffusion models (Kingma et al., 2021).

Individual differences in motivation can explain why people act differently in the same situation, and which aspects of a game people with different motive profiles may find most engaging. However, identifying a player’s motive profile from data available during gameplay remains an open research question. Besides a range of subjective and objective techniques for identifying player motivation, electroencephalography (EEG) technology could offer an automatic, objective technique for identifying the profile that best describes a given player. This article proposes a framework to measure player profiles of achievement, affiliation, and power motivation using EEG signals during their engagement within a game. First, an abstract mini-game is proposed to evaluate a player’s motivation. In the mini-game, each human player interacts with four nonplayer characters to gainfortuneorfriendshipthrough an individual play phase and a social network phase. The game is used within an experimental scenario to collect players’ actions and EEG signals. In addition, data from a psychological test are used to establish ground truth. We propose three subject labeling schemes using the output of the psychological test. Based on a player’s motive profile, behavioral indicators and EEG data analysis indicate that assessing a player’s motive profile is more robust from EEG signals than from behavioral data.

This paper proposes an image registration method to address the problem of high registration delay, object matching inaccuracy and even object loss caused by fast camera motion in mobile augmented reality (MAR). Firstly, TLD (Tracking Learning Detection algorithm) is improved by creating the background model for the feature points and their neighborhood to reduce the number of scanning windows and to accelerate object tracking. Then, an improved ORB (Oriented FAST and Rotated BRIEF algorithm) is presented, where Hessian matrix and gray centroid method are adopted to give the scale and direction information to the feature points, Hamming distance is used to match the feature points, and RANSAC (RANdom SAmple Consensus algorithm) further optimizes the matching results. Furthermore, the image registration algorithm based on improved TLD and improved ORB is proposed, which reduces the detection range of feature points, improves the speed of object tracking, and can quickly calculate the transformation matrix between virtual and real scenes. The experimental results show that the proposed image registration algorithm has high accuracy and good robustness in the mobile augmented reality, so it can better meet the application needs of augmented reality in the mobile terminal.

Individual head-related transfer functions (HRTFs) are necessary for rendering authentic spatial perceptions in spatial audio applications. To obtain individual HRTFs while avoiding tedious and complicated measurement and calculation, an improved customization method based on anthropometry matching is proposed. In the method, a set of HRTFs, which is the best match to the pinna shape of the listener using four pinna-related anatomical parameters, is selected as the listener's individual HRTFs from a pre-acquired HRTF baseline database. A series of subject localization experiments was conducted to verify the effectiveness of the proposed method compared with the existing method. Results show that the median-plane localization performance of the customization method proposed in the present work is prior to the existing method, though performance improvement varies with source position.