Weaknesses: The main weakness of this paper is that a very similar problem was considered in reference [31], which has a nearly identical title. The main difference between this paper and [31] seems to be in the method: this paper uses gradient tracking, while [31] does not. Nevertheless, both papers show convergence to a neighborhood of the optimal solution, so the theoretical innovation in this paper is not sufficient for NeurIPS publication. The authors mention that the empirical results of this paper are better, but this paper seems focused on theory with a relatively brief simulation section, so it should be evaluated as a theory paper. Additionally, the results here fall short of what one wants to achieve in this setting, which is convergence to the optimal solution.

This paper is theoretical work that provides finite time analysis for decentralised TD learning. The reviewers and myself, although not anonymously, think this contribution may be significant and interesting to the community due to recent interest in the finite time analysis of TD algorithms and (linear) function approximation. We request the authors to address the changes required in the manuscript. The authors propose a distributed method for safety. The reviewers and myself were not convinced that this paper proposes a novel method, specifically, due to lack of proper comparison to previous work.

Effective traffic signal control (TSC) is crucial in mitigating urban congestion and reducing emissions. Recently, reinforcement learning (RL) has been the research trend for TSC. However, existing RL algorithms face several real-world challenges that hinder their practical deployment in TSC: (1) Sensor accuracy deteriorates with increased sensor detection range, and data transmission is prone to noise, potentially resulting in unsafe TSC decisions. (2) During the training of online RL, interactions with the environment could be unstable, potentially leading to inappropriate traffic signal phase (TSP) selection and traffic congestion. (3) Most current TSC algorithms focus only on TSP decisions, overlooking the critical aspect of phase duration, affecting safety and efficiency. To overcome these challenges, we propose a robust two-stage fuzzy approach called FuzzyLight, which integrates compressed sensing and RL for TSC deployment. FuzzyLight offers several key contributions: (1) It employs fuzzy logic and compressed sensing to address sensor noise and enhances the efficiency of TSP decisions. (2) It maintains stable performance during training and combines fuzzy logic with RL to generate precise phases. (3) It works in real cities across 22 intersections and demonstrates superior performance in both real-world and simulated environments. Experimental results indicate that FuzzyLight enhances traffic efficiency by 48% compared to expert-designed timings in the real world. Furthermore, it achieves state-of-the-art (SOTA) performance in simulated environments using six real-world datasets with transmission noise. The code and deployment video are available at the URL1

Depending on their sophistication level, sensors can be classified ranging from simple sensors that directly measure single physical parameters (e.g., ambient light sensors and temperature sensors) to complex intelligent sensors, which determine parameters of the surrounding environment through wide spectrum signals (e.g., radio frequency/radar and light/video); besides measuring, they perform data processing and are enabled to carry out actuations. Whereas intelligent sensors make use of data of a different nature underneath, in which complex and nonlinear behaviors are codified; data-mining techniques used jointly with machine learning (ML) algorithms have shown adequate performance for modeling this hidden information. As intelligent sensors often rely on complex sensors and sensor fusion techniques, the data processing power they need can only be provided by high-performance computational platforms such as microprocessors, graphics-processing units (GPUs), or field-programmable gate arrays (FPGAs). In particular, FPGA-based implementations stand out due to the extremely high operational frequencies and low power consumption they can achieve, even for complex, multilayered algorithms [1]. In the context of the automotive field, intelligent sensors are key components of current assistance systems.

This paper deals with an important problem in theoretical reinforcement learning (RL), that is, finite-time analysis of on-policy RL algorithms such as SARSA. If the analysis techniques, as well as proofs, were correct and concrete, this work may have a broad impact on analyzing related stochastic approximation/RL algorithms. Although important and interesting, the present submission contains several major concerns, that have limited the contributions and even brought into question the practical usefulness of the reported theoretical results. These concerns are listed as follows. To facilitate analysis, a number of the assumptions adopted in this work are strong and impractical.

Because the initial reviews were mixed, I obtained an additional review from an expert in the area of this paper. This 4th review came back clearly positive, but in the mean time one of the positive reviewers changed to negative (and later one of the negatives turned to positive). Then we had a lot of discussion, but the reviewers never did agree on how best to view this paper. In fact, they seemed to talk past each other, and in the end we had two positive and two negative reviews. As the area chair, reading the reviews and listening to the discussion, I found the 4th, very-positive review to be the most compelling.

The paper targets the problem of measuring the representation power of Graph neural networks (GNNs), an interesting and important topic, that has become popular recently (partially due to two prominent works (Xu et al. There are three main contributions: 1. Establishing the equivalence between two methods for measuring GNN representation power: (i) their ability to approximate permutation invariant functions (ii) their ability to distinguish non-isomorphic graphs. Although not very surprising, this is a nice observation. The authors show that these sigma algebras are an equivalent way to measure representation power of GNNs, for instance, the inclusion of sigma algebras originating from two models is equivalent to saying one model is more powerful than the other. This is a potentially useful observation.

This paper leverages the graph isomorphism problem to study the expressive power of GNNs. In addition, a measure of expressiveness is formalized using sigma-algebras and the authors propose a novel variant of GNN, RING-GNN, that is evaluated in an experimental study where it shows competitive results. The reviewers agree that this is a nice contribution, the theoretical results are interesting (though somehow expected) and that the proposed extension of G-invariant networks is relevant. However, all reviewers agree that the experimental comparison with RING-GNN-SVD is unfair and MUST BE REMOVED in a published version of the paper (that is removing the last line from table 1). One of the reviewer also note that a comparison with LanczosNet should be included (though the lack of comparison is not ground for rejection).

Overall, the paper made significant contribution to both the reinforcement learning community and optimization community. The proposed algorithm is a variant of non-convex SAGA algorithm introduced by [1]. The novelty comes from their proof for the non-convex but strongly concave case. There are several issues which should be addressed: 1, Recasting the policy evaluation as a primal-dual optimization via the Fenchel duality technique is not new. In fact, [2,3,4] have already exploit this reformulation. First, these related work should be referred appropriately.

The main contribution of this paper is in solving the finite-sum minimax problem arising from off-line policy evaluation with nonlinear function approximation. The minimax problem is non-convex in the primal variable and strong convexity in the dual subproblem, and a single time-scale algorithm is proposed to find an approximate stationary point. Although it does not address the full stochastic TD learning problem, the progress in the finite-sum off-line version is quite meaningful.