Policy Optimization Algorithms in a Unified Framework

Policy optimization algorithms are crucial in many fields but challenging to grasp and implement, often due to complex calculations related to Markov decision processes and varying use of discount and average reward setups. This paper presents a unified framework that applies generalized ergodicity theory and perturbation analysis to clarify and enhance the application of these algorithms. Generalized ergodicity theory sheds light on the steady-state behavior of stochastic processes, aiding understanding of both discounted and average rewards. Perturbation analysis provides in-depth insights into the fundamental principles of policy optimization algorithms. We use this framework to identify common implementation errors and demonstrate the correct approaches. Through a case study on Linear Quadratic Regulator problems, we illustrate how slight variations in algorithm design affect implementation outcomes. We aim to make policy optimization algorithms more accessible and reduce their misuse in practice.

All Patches Matter, More Patches Better: Enhance AI-Generated Image Detection via Panoptic Patch Learning

The exponential growth of AI-generated images (AIGIs) underscores the urgent need for robust and generalizable detection methods. In this paper, we establish two key principles for AIGI detection through systematic analysis: \textbf{(1) All Patches Matter:} Unlike conventional image classification where discriminative features concentrate on object-centric regions, each patch in AIGIs inherently contains synthetic artifacts due to the uniform generation process, suggesting that every patch serves as an important artifact source for detection. \textbf{(2) More Patches Better}: Leveraging distributed artifacts across more patches improves detection robustness by capturing complementary forensic evidence and reducing over-reliance on specific patches, thereby enhancing robustness and generalization. However, our counterfactual analysis reveals an undesirable phenomenon: naively trained detectors often exhibit a \textbf{Few-Patch Bias}, discriminating between real and synthetic images based on minority patches. We identify \textbf{Lazy Learner} as the root cause: detectors preferentially learn conspicuous artifacts in limited patches while neglecting broader artifact distributions. To address this bias, we propose the \textbf{P}anoptic \textbf{P}atch \textbf{L}earning (PPL) framework, involving: (1) Random Patch Replacement that randomly substitutes synthetic patches with real counterparts to compel models to identify artifacts in underutilized regions, encouraging the broader use of more patches; (2) Patch-wise Contrastive Learning that enforces consistent discriminative capability across all patches, ensuring uniform utilization of all patches. Extensive experiments across two different settings on several benchmarks verify the effectiveness of our approach.

Frequency-Aware Attention-LSTM for PM$_{2.5}$ Time Series Forecasting

To enhance the accuracy and robustness of PM$_{2.5}$ concentration forecasting, this paper introduces FALNet, a Frequency-Aware LSTM Network that integrates frequency-domain decomposition, temporal modeling, and attention-based refinement. The model first applies STL and FFT to extract trend, seasonal, and denoised residual components, effectively filtering out high-frequency noise. The filtered residuals are then fed into a stacked LSTM to capture long-term dependencies, followed by a multi-head attention mechanism that dynamically focuses on key time steps. Experiments conducted on real-world urban air quality datasets demonstrate that FALNet consistently outperforms conventional models across standard metrics such as MAE, RMSE, and $R^2$. The model shows strong adaptability in capturing sharp fluctuations during pollution peaks and non-stationary conditions. These results validate the effectiveness and generalizability of FALNet for real-time air pollution prediction, environmental risk assessment, and decision-making support.

SpatialCoT: Advancing Spatial Reasoning through Coordinate Alignment and Chain-of-Thought for Embodied Task Planning

Spatial reasoning is an essential problem in embodied AI research. Efforts to enhance spatial reasoning abilities through supplementary spatial data and fine-tuning have proven limited and ineffective when addressing complex embodied tasks, largely due to their dependence on language-based outputs. While some approaches have introduced a point-based action space to mitigate this issue, they fall short in managing more intricate tasks within complex environments. This deficiency arises from their failure to fully exploit the inherent thinking and reasoning capabilities that are fundamental strengths of Vision-Language Models (VLMs). To address these limitations, we propose a novel approach named SpatialCoT, specifically designed to bolster the spatial reasoning capabilities of VLMs. Our approach comprises two stages: spatial coordinate bi-directional alignment, which aligns vision-language inputs with spatial coordinates, and chain-of-thought spatial grounding, which harnesses the reasoning capabilities of language models for advanced spatial reasoning. We evaluate SpatialCoT on challenging navigation and manipulation tasks, both in simulation and real-world settings. Experimental results demonstrate that our method significantly outperforms previous state-of-the-art approaches in both tasks.

Action-Attentive Deep Reinforcement Learning for Autonomous Alignment of Beamlines

Synchrotron radiation sources play a crucial role in fields such as materials science, biology, and chemistry. The beamline, a key subsystem of the synchrotron, modulates and directs the radiation to the sample for analysis. However, the alignment of beamlines is a complex and time-consuming process, primarily carried out manually by experienced engineers. Even minor misalignments in optical components can significantly affect the beam's properties, leading to suboptimal experimental outcomes. Current automated methods, such as bayesian optimization (BO) and reinforcement learning (RL), although these methods enhance performance, limitations remain. The relationship between the current and target beam properties, crucial for determining the adjustment, is not fully considered. Additionally, the physical characteristics of optical elements are overlooked, such as the need to adjust specific devices to control the output beam's spot size or position. This paper addresses the alignment of beamlines by modeling it as a Markov Decision Process (MDP) and training an intelligent agent using RL. The agent calculates adjustment values based on the current and target beam states, executes actions, and iterates until optimal parameters are achieved. A policy network with action attention is designed to improve decision-making by considering both state differences and the impact of optical components. Experiments on two simulated beamlines demonstrate that our algorithm outperforms existing methods, with ablation studies highlighting the effectiveness of the action attention-based policy network.

MARS: Unleashing the Power of Variance Reduction for Training Large Models

Training deep neural networks--and more recently, large models--demands efficient and scalable optimizers. Adaptive gradient algorithms like Adam, AdamW, and their variants have been central to this task. Despite the development of numerous variance reduction algorithms in the past decade aimed at accelerating stochastic optimization in both convex and nonconvex settings, variance reduction has not found widespread success in training deep neural networks or large language models. Consequently, it has remained a less favored approach in modern AI. In this paper, to unleash the power of variance reduction for efficient training of large models, we propose a unified optimization framework, MARS (Make vAriance Reduction Shine), which reconciles preconditioned gradient methods with variance reduction via a scaled stochastic recursive momentum technique. Within our framework, we introduce three instances of MARS that leverage preconditioned gradient updates based on AdamW, Lion, and Shampoo, respectively. We also draw a connection between our algorithms and existing optimizers. Experimental results on training GPT-2 models indicate that MARS consistently outperforms AdamW by a large margin.

Decoupled Data Augmentation for Improving Image Classification

Recent advancements in image mixing and generative data augmentation have shown promise in enhancing image classification. However, these techniques face the challenge of balancing semantic fidelity with diversity. Specifically, image mixing involves interpolating two images to create a new one, but this pixel-level interpolation can compromise fidelity. Generative augmentation uses text-to-image generative models to synthesize or modify images, often limiting diversity to avoid generating out-of-distribution data that potentially affects accuracy. We propose that this fidelity-diversity dilemma partially stems from the whole-image paradigm of existing methods. Since an image comprises the class-dependent part (CDP) and the class-independent part (CIP), where each part has fundamentally different impacts on the image's fidelity, treating different parts uniformly can therefore be misleading. To address this fidelity-diversity dilemma, we introduce Decoupled Data Augmentation (De-DA), which resolves the dilemma by separating images into CDPs and CIPs and handling them adaptively. To maintain fidelity, we use generative models to modify real CDPs under controlled conditions, preserving semantic consistency. To enhance diversity, we replace the image's CIP with inter-class variants, creating diverse CDP-CIP combinations. Additionally, we implement an online randomized combination strategy during training to generate numerous distinct CDP-CIP combinations cost-effectively. Comprehensive empirical evaluations validate the effectiveness of our method.

UniVoxel: Fast Inverse Rendering by Unified Voxelization of Scene Representation

Typical inverse rendering methods focus on learning implicit neural scene representations by modeling the geometry, materials and illumination separately, which entails significant computations for optimization. In this work we design a Unified Voxelization framework for explicit learning of scene representations, dubbed UniVoxel, which allows for efficient modeling of the geometry, materials and illumination jointly, thereby accelerating the inverse rendering significantly. To be specific, we propose to encode a scene into a latent volumetric representation, based on which the geometry, materials and illumination can be readily learned via lightweight neural networks in a unified manner. Particularly, an essential design of UniVoxel is that we leverage local Spherical Gaussians to represent the incident light radiance, which enables the seamless integration of modeling illumination into the unified voxelization framework. Such novel design enables our UniVoxel to model the joint effects of direct lighting, indirect lighting and light visibility efficiently without expensive multi-bounce ray tracing. Extensive experiments on multiple benchmarks covering diverse scenes demonstrate that UniVoxel boosts the optimization efficiency significantly compared to other methods, reducing the per-scene training time from hours to 18 minutes, while achieving favorable reconstruction quality. Code is available at https://github.com/freemantom/UniVoxel.

DiffExplainer: Unveiling Black Box Models Via Counterfactual Generation

In the field of medical imaging, particularly in tasks related to early disease detection and prognosis, understanding the reasoning behind AI model predictions is imperative for assessing their reliability. Conventional explanation methods encounter challenges in identifying decisive features in medical image classifications, especially when discriminative features are subtle or not immediately evident. To address this limitation, we propose an agent model capable of generating counterfactual images that prompt different decisions when plugged into a black box model. By employing this agent model, we can uncover influential image patterns that impact the black model's final predictions. Through our methodology, we efficiently identify features that influence decisions of the deep black box. We validated our approach in the rigorous domain of medical prognosis tasks, showcasing its efficacy and potential to enhance the reliability of deep learning models in medical image classification compared to existing interpretation methods. The code will be publicly available at https://github.com/ayanglab/DiffExplainer.

Graph Neural Thompson Sampling

We consider an online decision-making problem with a reward function defined over graph-structured data. We formally formulate the problem as an instance of graph action bandit. We then propose \texttt{GNN-TS}, a Graph Neural Network (GNN) powered Thompson Sampling (TS) algorithm which employs a GNN approximator for estimating the mean reward function and the graph neural tangent features for uncertainty estimation. We prove that, under certain boundness assumptions on the reward function, GNN-TS achieves a state-of-the-art regret bound which is (1) sub-linear of order $\tilde{\mathcal{O}}((\tilde{d} T)^{1/2})$ in the number of interaction rounds, $T$, and a notion of effective dimension $\tilde{d}$, and (2) independent of the number of graph nodes. Empirical results validate that our proposed \texttt{GNN-TS} exhibits competitive performance and scales well on graph action bandit problems.