Search

We introduce Centaurus, a class of networks composed of generalized state-space model (SSM) blocks, where the SSM operations can be treated as tensor contractions during training. The optimal order of tensor contractions can then be systematically determined for every SSM block to maximize training efficiency. This allows more flexibility in designing SSM blocks beyond the depthwise-separable configuration commonly implemented. The new design choices will take inspiration from classical convolutional blocks including group convolutions, full convolutions, and bottleneck blocks. We architect the Centaurus network with a mixture of these blocks, to balance between network size and performance, as well as memory and computational efficiency during both training and inference. We show that this heterogeneous network design outperforms its homogeneous counterparts in raw audio processing tasks including keyword spotting, speech denoising, and automatic speech recognition (ASR). For ASR, Centaurus is the first network with competitive performance that can be made fully state-space based, without using any nonlinear recurrence (LSTMs), explicit convolutions (CNNs), or (surrogate) attention mechanism. Source code is available at github.com/Brainchip-Inc/Centaurus, Comment: 25 pages, 7 figures

The distributed recursion (DR) algorithm is an effective method for solving the pooling problem that arises in many applications. It is based on the well-known P-formulation of the pooling problem, which involves the flow and quality variables; and it can be seen as a variant of the successive linear programming (SLP) algorithm, where the linear programming (LP) approximation problem can be transformed from the LP approximation problem derived by using the first-order Taylor series expansion technique. In this paper, we first propose a new nonlinear programming (NLP) formulation for the pooling problem involving only the flow variables, and show that the DR algorithm can be seen as a direct application of the SLP algorithm to the newly proposed formulation. With this new useful theoretical insight, we then develop a new variant of DR algorithm, called penalty DR (PDR) algorithm, based on the proposed formulation. The proposed PDR algorithm is a penalty algorithm where violations of the (linearized) nonlinear constraints are penalized in the objective function of the LP approximation problem with the penalty terms increasing when the constraint violations tend to be large. Compared with the LP approximation problem in the classic DR algorithm, the LP approximation problem in the proposed PDR algorithm can return a solution with a better objective value, which makes it more suitable for finding high-quality solutions for the pooling problem. Numerical experiments on benchmark and randomly constructed instances show that the proposed PDR algorithm is more effective than the classic SLP and DR algorithms in terms of finding a better solution for the pooling problem., Comment: 22 pages, 2 figures, submitted for possible publication

MuZero has achieved superhuman performance in various games by using a dynamics network to predict environment dynamics for planning, without relying on simulators. However, the latent states learned by the dynamics network make its planning process opaque. This paper aims to demystify MuZero's model by interpreting the learned latent states. We incorporate observation reconstruction and state consistency into MuZero training and conduct an in-depth analysis to evaluate latent states across two board games: 9x9 Go and Outer-Open Gomoku, and three Atari games: Breakout, Ms. Pacman, and Pong. Our findings reveal that while the dynamics network becomes less accurate over longer simulations, MuZero still performs effectively by using planning to correct errors. Our experiments also show that the dynamics network learns better latent states in board games than in Atari games. These insights contribute to a better understanding of MuZero and offer directions for future research to improve the playing performance, robustness, and interpretability of the MuZero algorithm., Comment: Accepted by the 29th International Conference on Technologies and Applications of Artificial Intelligence (TAAI 2024)

Although AlphaZero has achieved superhuman levels in Go, recent research has highlighted its vulnerability in particular situations requiring a more comprehensive understanding of the entire board. To address this challenge, this paper introduces ResTNet, a network that interleaves residual networks and Transformer. Our empirical experiments demonstrate several advantages of using ResTNet. First, it not only improves playing strength but also enhances the ability of global information. Second, it defends against an adversary Go program, called cyclic-adversary, tailor-made for attacking AlphaZero algorithms, significantly reducing the average probability of being attacked rate from 70.44% to 23.91%. Third, it improves the accuracy from 59.15% to 80.01% in correctly recognizing ladder patterns, which are one of the challenging patterns for Go AIs. Finally, ResTNet offers a potential explanation of the decision-making process and can also be applied to other games like Hex. To the best of our knowledge, ResTNet is the first to integrate residual networks and Transformer in the context of AlphaZero for board games, suggesting a promising direction for enhancing AlphaZero's global understanding.

We present aTENNuate, a simple deep state-space autoencoder configured for efficient online raw speech enhancement in an end-to-end fashion. The network's performance is primarily evaluated on raw speech denoising, with additional assessments on tasks such as super-resolution and de-quantization. We benchmark aTENNuate on the VoiceBank + DEMAND and the Microsoft DNS1 synthetic test sets. The network outperforms previous real-time denoising models in terms of PESQ score, parameter count, MACs, and latency. Even as a raw waveform processing model, the model maintains high fidelity to the clean signal with minimal audible artifacts. In addition, the model remains performant even when the noisy input is compressed down to 4000Hz and 4 bits, suggesting general speech enhancement capabilities in low-resource environments. Code is available at github.com/Brainchip-Inc/aTENNuate

In the paper, we consider the competitive facility location problem with limited choice rule (CFLPLCR), which attempts to open a subset of facilities to maximize the net profit of a newcomer company, requiring customers to patronize only a limited number of opening facilities and an outside option. We propose an efficient branch-and-cut (B&C) approach for the CFLPLCR based on newly proposed mixed integer linear programming (MILP) formulations. Specifically, by establishing the submodularity of the probability function, we develop an MILP formulation for the CFLPLCR using the submodular inequalities. For the special case where each customer patronizes at most one open facility and the outside option, we show that the submodular inequalities can characterize the convex hull of the considered set and provide a compact MILP formulation. Moreover, for the general case, we strengthen the submodular inequalities by sequential lifting, resulting in a class of facet-defining inequalities. The proposed lifted submodular inequalities are shown to be stronger than the classic submodular inequalities, enabling to obtain another MILP formulation with a tighter linear programming (LP) relaxation. By extensive numerical experiments, we show that the proposed B&C approach outperforms the state-of-the-art generalized Benders decomposition approach by at least one order of magnitude. Furthermore, it enables to solve CFLPLCR instances with 10000 customers and 2000 facilities., Comment: 29 pages, 2 figures, submitted for possible publication

We introduce a neural network named PLEIADES (PoLynomial Expansion In Adaptive Distributed Event-based Systems), belonging to the TENNs (Temporal Neural Networks) architecture. We focus on interfacing these networks with event-based data to perform online spatiotemporal classification and detection with low latency. By virtue of using structured temporal kernels and event-based data, we have the freedom to vary the sample rate of the data along with the discretization step-size of the network without additional finetuning. We experimented with three event-based benchmarks and obtained state-of-the-art results on all three by large margins with significantly smaller memory and compute costs. We achieved: 1) 99.59% accuracy with 192K parameters on the DVS128 hand gesture recognition dataset and 100% with a small additional output filter; 2) 99.58% test accuracy with 277K parameters on the AIS 2024 eye tracking challenge; and 3) 0.556 mAP with 576k parameters on the PROPHESEE 1 Megapixel Automotive Detection Dataset., Comment: 11 pages, 3 figures

This survey reviews the AIS 2024 Event-Based Eye Tracking (EET) Challenge. The task of the challenge focuses on processing eye movement recorded with event cameras and predicting the pupil center of the eye. The challenge emphasizes efficient eye tracking with event cameras to achieve good task accuracy and efficiency trade-off. During the challenge period, 38 participants registered for the Kaggle competition, and 8 teams submitted a challenge factsheet. The novel and diverse methods from the submitted factsheets are reviewed and analyzed in this survey to advance future event-based eye tracking research., Comment: Qinyu Chen is the corresponding author

Event-based data are commonly encountered in edge computing environments where efficiency and low latency are critical. To interface with such data and leverage their rich temporal features, we propose a causal spatiotemporal convolutional network. This solution targets efficient implementation on edge-appropriate hardware with limited resources in three ways: 1) deliberately targets a simple architecture and set of operations (convolutions, ReLU activations) 2) can be configured to perform online inference efficiently via buffering of layer outputs 3) can achieve more than 90% activation sparsity through regularization during training, enabling very significant efficiency gains on event-based processors. In addition, we propose a general affine augmentation strategy acting directly on the events, which alleviates the problem of dataset scarcity for event-based systems. We apply our model on the AIS 2024 event-based eye tracking challenge, reaching a score of 0.9916 p10 accuracy on the Kaggle private testset., Comment: 8 pages, 3 figures

Effects of tensor spin polarization (TSP) on the chiral restoration and deconfinement phase transitions are studied in Polyakov loop extended Nambu-Jona-Lasinio (PNJL) model. For chiral phase transition, the higher the polarized degree of quark-antiquark pairs under the strong magnetic field, the higher the phase transition temperature. The TSP corrects the position of the critical end point. The small impact of TSP on the phase transition temperature is found for the deconfinement phase transition. On the other hand, we divide the phase space into three ranges based on the phase diagram obtained from the PNJL model: the confinement phase with chiral symmetry broken, the deconfinement phase with restored chiral symmetry, and the confinement phase with restored chiral symmetry (quarkyonic phase). It is found that TSP has only a very small effect on the anisotropic pressure in the deconfined phase with chiral symmetry restored and the quarkyonic phase, but it has a very strong effect on the anisotropic pressure in the confined phase with chiral symmetry broken. This is because TSP is closely related to chiral symmetry. The restoration of chiral symmetry means the dissociation of spin polarization condensate., Comment: 20 pages, 7 figures