arXiv subjects

Yi Ma

Publications and source records attributed to Yi Ma.

At least 73 records · Page 4Linked to original sources

arXiv2025

GAUSS: Benchmarking Structured Mathematical Skills for Large Language Models

We introduce \textbf{GAUSS} (\textbf{G}eneral \textbf{A}ssessment of \textbf{U}nderlying \textbf{S}tructured \textbf{S}kills in Mathematics), a benchmark that evaluates LLMs' mathematical abilities across twelve core skill dimensions, grouped into three domains: knowledge and understanding, problem solving and communication, and meta-skills and creativity. By categorizing problems according to cognitive skills and designing tasks that isolate specific abilities, GAUSS constructs comprehensive, fine-grained, and interpretable profiles of models' mathematical abilities. These profiles faithfully represent their underlying mathematical intelligence. To exemplify how to use the \textsc{GAUSS} benchmark, we have derived the skill profile of \textsc{GPT-5-thinking}, revealing its strengths and weaknesses as well as its differences relative to \textsc{o4-mini-high}, thereby underscoring the value of multidimensional, skill-based evaluation.

cs.AI↗

arXiv2025

On the Edge of Memorization in Diffusion Models

When do diffusion models reproduce their training data, and when are they able to generate samples beyond it? A practically relevant theoretical understanding of this interplay between memorization and generalization may significantly impact real-world deployments of diffusion models with respect to issues such as copyright infringement and data privacy. In this work, to disentangle the different factors that influence memorization and generalization in practical diffusion models, we introduce a scientific and mathematical "laboratory" for investigating these phenomena in diffusion models trained on fully synthetic or natural image-like structured data. Within this setting, we hypothesize that the memorization or generalization behavior of an underparameterized trained model is determined by the difference in training loss between an associated memorizing model and a generalizing model. To probe this hypothesis, we theoretically characterize a crossover point wherein the weighted training loss of a fully generalizing model becomes greater than that of an underparameterized memorizing model at a critical value of model (under)parameterization. We then demonstrate via carefully-designed experiments that the location of this crossover predicts a phase transition in diffusion models trained via gradient descent, validating our hypothesis. Ultimately, our theory enables us to analytically predict the model size at which memorization becomes predominant. Our work provides an analytically tractable and practically meaningful setting for future theoretical and empirical investigations. Code for our experiments is available at https://github.com/DruvPai/diffusion_mem_gen.

cs.LG↗

arXiv2025

Data-Importance-Aware Power Allocation for Adaptive Semantic Communication in Computer Vision Applications

Life-transformative applications such as immersive extended reality are revolutionizing wireless communications and computer vision (CV). This paper presents a novel framework for importance-aware adaptive data transmissions, designed specifically for real-time CV applications where task-specific fidelity is critical. A novel importance-weighted mean square error (IMSE) metric is introduced as a task-oriented measure of reconstruction quality, considering sub-pixel-level importance (SP-I) and semantic segment-level importance (SS-I) models. To minimize IMSE under total power constraints, data-importance-aware waterfilling approaches are proposed to optimally allocate transmission power according to data importance and channel conditions, prioritizing sub-streams with high importance. Simulation results demonstrate that the proposed approaches significantly outperform margin-adaptive waterfilling and equal power allocation strategies. The data partitioning that combines both SP-I and SS-I models is shown to achieve the most significant improvements, with normalized IMSE gains exceeding $7\,$dB and $10\,$dB over the baselines at high SNRs ($>10\,$dB). These substantial gains highlight the potential of the proposed framework to enhance data efficiency and robustness in real-time CV applications, especially in bandwidth-limited and resource-constrained environments.

eess.SP↗

arXiv2025

Scaling DRL for Decision Making: A Survey on Data, Network, and Training Budget Strategies

In recent years, the expansion of neural network models and training data has driven remarkable progress in deep learning, particularly in computer vision and natural language processing. This advancement is underpinned by the concept of Scaling Laws, which demonstrates that scaling model parameters and training data enhances learning performance. While these fields have witnessed breakthroughs, such as the development of large language models like GPT-4 and advanced vision models like Midjourney, the application of scaling laws in deep reinforcement learning (DRL) remains relatively unexplored. Despite its potential to improve performance, the integration of scaling laws into DRL for decision making has not been fully realized. This review addresses this gap by systematically analyzing scaling strategies in three dimensions: data, network, and training budget. In data scaling, we explore methods to optimize data efficiency through parallel sampling and data generation, examining the relationship between data volume and learning outcomes. For network scaling, we investigate architectural enhancements, including monolithic expansions, ensemble and MoE methods, and agent number scaling techniques, which collectively enhance model expressivity while posing unique computational challenges. Lastly, in training budget scaling, we evaluate the impact of distributed training, high replay ratios, large batch sizes, and auxiliary training on training efficiency and convergence. By synthesizing these strategies, this review not only highlights their synergistic roles in advancing DRL for decision making but also provides a roadmap for future research. We emphasize the importance of balancing scalability with computational efficiency and outline promising directions for leveraging scaling to unlock the full potential of DRL in various tasks such as robot control, autonomous driving and LLM training.

cs.LG↗

arXiv2025

CAD-MLLM: Unifying Multimodality-Conditioned CAD Generation With MLLM

This paper aims to design a unified Computer-Aided Design (CAD) generation system that can easily generate CAD models based on the user's inputs in the form of textual description, images, point clouds, or even a combination of them. Towards this goal, we introduce the CAD-MLLM, the first system capable of generating parametric CAD models conditioned on the multimodal input. Specifically, within the CAD-MLLM framework, we leverage the command sequences of CAD models and then employ advanced large language models (LLMs) to align the feature space across these diverse multi-modalities data and CAD models' vectorized representations. To facilitate the model training, we design a comprehensive data construction and annotation pipeline that equips each CAD model with corresponding multimodal data. Our resulting dataset, named Omni-CAD, is the first multimodal CAD dataset that contains textual description, multi-view images, points, and command sequence for each CAD model. It contains approximately 450K instances and their CAD construction sequences. To thoroughly evaluate the quality of our generated CAD models, we go beyond current evaluation metrics that focus on reconstruction quality by introducing additional metrics that assess topology quality and surface enclosure extent. Extensive experimental results demonstrate that CAD-MLLM significantly outperforms existing conditional generative methods and remains highly robust to noises and missing points. The project page and more visualizations can be found at: https://cad-mllm.github.io/

cs.CV↗

arXiv2025

LO-Aware Adaptive Modulation for Rydberg Atomic Receivers

Rydberg atomic (RA) receivers represent a revolutionary quantum technology for wireless communications, offering unprecedented sensitivity beyond conventional radio frequency (RF) antennas. However, these receivers detect only signal amplitude, losing critical phase information. While reference signals generated by a local oscillator (LO) can assist in phase recovery, existing modulation schemes designed for conventional systems perform poorly with this quantum detection mechanism. This paper introduces a breakthrough LO-aware adaptive modulation (LOAM) scheme specifically developed for RA receivers that dynamically adapts to complex fading channel coefficients. LOAM maximizes the minimum amplitude difference between constellation points, ensuring optimal detection performance. The innovation employs an adaptive co-linear constellation architecture aligned with the combined phase of reference signal and channel coefficient. For strong reference signals, LOAM generates symmetric constellation points centered at origin; for weak signals, it adopts non-symmetric distributions. The paper mathematically derives the threshold governing these operational regimes. Simulation results reveal the transformative impact of LOAM, demonstrating performance gains exceeding 45 dB over conventional modulation schemes, including quadrature amplitude modulation (QAM), phase-shift keying (PSK), and pulse-amplitude modulation (PAM).

cs.IT↗

arXiv2025

Deep Learning-Based Rate-Adaptive CSI Feedback for Wideband XL-MIMO Systems in the Near-Field Domain

Accurate and efficient channel state information (CSI) feedback is crucial for unlocking the substantial spectral efficiency gains of extremely large-scale MIMO (XL-MIMO) systems in future 6G networks. However, the combination of near-field spherical wave propagation and frequency-dependent beam split effects in wideband scenarios poses significant challenges for CSI representation and compression. This paper proposes WideNLNet-CA, a rate-adaptive deep learning framework designed to enable efficient CSI feedback in wideband near-field XL-MIMO systems. WideNLNet-CA introduces a lightweight encoder-decoder architecture with multi-stage downsampling and upsampling, incorporating computationally efficient residual blocks to capture complex multi-scale channel features with reduced overhead. A novel compression ratio adaptive module with feature importance estimation is introduced to dynamically modulate feature selection based on target compression ratios, enabling flexible adaptation across a wide range of feedback rates using a single model. Evaluation results demonstrate that WideNLNet-CA consistently outperforms existing compressive sensing and deep learning-based works across various compression ratios and bandwidths, while maintaining fast inference and low model storage requirements.

cs.IT↗

arXiv2025

Viser: Imperative, Web-based 3D Visualization in Python

We present Viser, a 3D visualization library for computer vision and robotics. Viser aims to bring easy and extensible 3D visualization to Python: we provide a comprehensive set of 3D scene and 2D GUI primitives, which can be used independently with minimal setup or composed to build specialized interfaces. This technical report describes Viser's features, interface, and implementation. Key design choices include an imperative-style API and a web-based viewer, which improve compatibility with modern programming patterns and workflows.

cs.CV↗

arXiv2025

LightCom: A Generative AI-Augmented Framework for QoE-Oriented Communications

Data-intensive and immersive applications, such as virtual reality, impose stringent quality of experience (QoE) requirements that challenge traditional quality of service (QoS)-driven communication systems. This paper presents LightCom, a lightweight encoding and generative AI (GenAI)-augmented decoding framework, designed for QoE-oriented communications under low signal-to-noise ratio (SNR) conditions. LightCom simplifies transmitter design by applying basic low-pass filtering for source coding and minimal channel coding, significantly reducing processing complexity and energy consumption. At the receiver, GenAI models reconstruct high-fidelity content from highly compressed and degraded signals by leveraging generative priors to infer semantic and structural information beyond traditional decoding capabilities. The key design principles are analyzed, along with the sufficiency and error-resilience of the source representation. We also develop importance-aware power allocation strategies to enhance QoE and extend perceived coverage. Simulation results demonstrate that LightCom achieves up to a $14$ dB improvement in robustness and a $9$ dB gain in perceived coverage, outperforming traditional QoS-driven systems relying on sophisticated source and channel coding. This paradigm shift moves communication systems towards human-centric QoE metrics rather than bit-level fidelity, paving the way for more efficient and resilient wireless networks.

eess.SP↗

arXiv2025

Leveraging Bi-Directional Channel Reciprocity for Robust Ultra-Low-Rate Implicit CSI Feedback with Deep Learning

Deep learning-based implicit channel state information (CSI) feedback has been introduced to enhance spectral efficiency in massive MIMO systems. Existing methods often show performance degradation in ultra-low-rate scenarios and inadaptability across diverse environments. In this paper, we propose Dual-ImRUNet, an efficient uplink-assisted deep implicit CSI feedback framework incorporating two novel plug-in preprocessing modules to achieve ultra-low feedback rates while maintaining high environmental robustness. First, a novel bi-directional correlation enhancement module is proposed to strengthen the correlation between uplink and downlink CSI eigenvector matrices. This module projects highly correlated uplink and downlink channel matrices into their respective eigenspaces, effectively reducing redundancy for ultra-low-rate feedback. Second, an innovative input format alignment module is designed to maintain consistent data distributions at both encoder and decoder sides without extra transmission overhead, thereby enhancing robustness against environmental variations. Finally, we develop an efficient transformer-based implicit CSI feedback network to exploit angular-delay domain sparsity and bi-directional correlation for ultra-low-rate CSI compression. Simulation results demonstrate successful reduction of the feedback overhead by 85% compared with the state-of-the-art method and robustness against unseen environments.

eess.SP↗

arXiv2025

Squeeze the Soaked Sponge: Efficient Off-policy Reinforcement Finetuning for Large Language Model

Reinforcement Learning (RL) has demonstrated its potential to improve the reasoning ability of Large Language Models (LLMs). One major limitation of most existing Reinforcement Finetuning (RFT) methods is that they are on-policy RL in nature, i.e., data generated during the past learning process is not fully utilized. This inevitably comes at a significant cost of compute and time, posing a stringent bottleneck on continuing economic and efficient scaling. To this end, we launch the renaissance of off-policy RL and propose Reincarnating Mix-policy Proximal Policy Gradient (ReMix), a general approach to enable on-policy RFT methods like PPO and GRPO to leverage off-policy data. ReMix consists of three major components: (1) Mix-policy proximal policy gradient with an increased Update-To-Data (UTD) ratio for efficient training; (2) KL-Convex policy constraint to balance the trade-off between stability and flexibility; (3) Policy reincarnation to achieve a seamless transition from efficient early-stage learning to steady asymptotic improvement. In our experiments, we train a series of ReMix models upon PPO, GRPO and 1.5B, 7B base models. ReMix shows an average Pass@1 accuracy of 52.10% (for 1.5B model) with 0.079M response rollouts, 350 training steps and achieves 63.27%/64.39% (for 7B model) with 0.007M/0.011M response rollouts, 50/75 training steps, on five math reasoning benchmarks (i.e., AIME'24, AMC'23, Minerva, OlympiadBench, and MATH500). Compared with 15 recent advanced models, ReMix shows SOTA-level performance with an over 30x to 450x reduction in training cost in terms of rollout data volume. In addition, we reveal insightful findings via multifaceted analysis, including the implicit preference for shorter responses due to the Whipping Effect of off-policy discrepancy, the collapse mode of self-reflection behavior under the presence of severe off-policyness, etc.

cs.LG↗

arXiv2025

Beyond Simple Edits: X-Planner for Complex Instruction-Based Image Editing

Recent diffusion-based image editing methods have significantly advanced text-guided tasks but often struggle to interpret complex, indirect instructions. Moreover, current models frequently suffer from poor identity preservation, unintended edits, or rely heavily on manual masks. To address these challenges, we introduce X-Planner, a Multimodal Large Language Model (MLLM)-based planning system that effectively bridges user intent with editing model capabilities. X-Planner employs chain-of-thought reasoning to systematically decompose complex instructions into simpler, clear sub-instructions. For each sub-instruction, X-Planner automatically generates precise edit types and segmentation masks, eliminating manual intervention and ensuring localized, identity-preserving edits. Additionally, we propose a novel automated pipeline for generating large-scale data to train X-Planner which achieves state-of-the-art results on both existing benchmarks and our newly introduced complex editing benchmark.

cs.CV↗

arXiv2025

Hybrid Constellation Modulation for Symbol-Level Precoding in RIS-Enhanced MU-MISO Systems

The application of symbol-level precoding (SLP) in reconfigurable intelligent surfaces (RIS) enhanced multi-user multiple-input single-output (MU-MISO) systems faces two main challenges. First, the state-of-the-art joint reflecting and SLP optimization approach requires exhaustive enumeration of all possible transmit symbol combinations, resulting in scalability issues as the modulation order and number of users increase. Second, conventional quadrature amplitude modulation (QAM) exhibits strict constructive interference (CI) regions, limiting its effectiveness for CI exploitation in SLP. To address these challenges, this paper proposes a novel modulation scheme, termed hybrid-constellation modulation (HCM), which has a structure of superposed QAM and ASK sub-constellations (SCs). HCM extends the CI regions compared to QAM. Additionally, a two-stage reflecting and SLP optimization method is developed to support HCM. The proposed methods are designed for practical RIS with discrete phase shifts and has good scalability. Simulation results show that HCM achieves up to 1.5 dB and 1 dB SER gains over QAM with modulation order 16 and 64, respectively.

eess.SP↗

arXiv2025

Cluster-Aware Two-Stage Method for Fast Iterative MIMO Detection in LEO Satellite Communications

In this paper, a cluster-aware two-stage multiple-input multiple-output (MIMO) detection method is proposed for direct-to-cell satellite communications. The method achieves computational efficiency by exploiting a distinctive property of satellite MIMO channels: users within the same geographical cluster exhibit highly correlated channel characteristics due to their physical proximity, which typically impedes convergence in conventional iterative MIMO detectors. The proposed method implements a two-stage strategy that first eliminates intra-cluster interference using computationally efficient small matrix inversions, then utilizes these pre-computed matrices to accelerate standard iterative MIMO detectors such as Gauss-Seidel (GS) and symmetric successive over-relaxation (SSOR) for effective inter-cluster interference cancellation. Computer simulations demonstrate that the proposed method achieves more than 12 times faster convergence under perfect channel state information. Even when accounting for channel estimation errors, the method maintains 9 times faster convergence, demonstrating its robustness and effectiveness for next-generation satellite MIMO communications.

cs.IT↗

arXiv2025

Attention-Only Transformers via Unrolled Subspace Denoising

Despite the popularity of transformers in practice, their architectures are empirically designed and neither mathematically justified nor interpretable. Moreover, as indicated by many empirical studies, some components of transformer architectures may be redundant. To derive a fully interpretable transformer architecture with only necessary components, we contend that the goal of representation learning is to compress a set of noisy initial token representations towards a mixture of low-dimensional subspaces. To compress these noisy token representations, an associated denoising operation naturally takes the form of a multi-head (subspace) self-attention. By unrolling such iterative denoising operations into a deep network, we arrive at a highly compact architecture that consists of \textit{only} self-attention operators with skip connections at each layer. Moreover, we show that each layer performs highly efficient denoising: it improves the signal-to-noise ratio of token representations \textit{at a linear rate} with respect to the number of layers. Despite its simplicity, extensive experiments on vision and language tasks demonstrate that such a transformer achieves performance close to that of standard transformer architectures such as GPT-2 and CRATE.

cs.LG↗

arXiv2025

Language-Image Alignment with Fixed Text Encoders

Currently, the most dominant approach to establishing language-image alignment is to pre-train text and image encoders jointly through contrastive learning, such as CLIP and its variants. In this work, we question whether such a costly joint training is necessary. In particular, we investigate if a pre-trained fixed large language model (LLM) offers a good enough text encoder to guide visual representation learning. That is, we propose to learn Language-Image alignment with a Fixed Text encoder (LIFT) from an LLM by training only the image encoder. Somewhat surprisingly, through comprehensive benchmarking and ablation studies, we find that this much simplified framework LIFT is highly effective and it outperforms CLIP in most scenarios that involve compositional understanding and long captions, while achieving considerable gains in computational efficiency. Our work takes a first step towards systematically exploring how text embeddings from LLMs can guide visual learning and suggests an alternative design choice for learning language-aligned visual representations.

cs.CV↗

arXiv2025

Dynamic Resource Allocation in Distributed MIMO-LEO Satellite Networks

This paper characterizes the impacts of channel estimation errors and Rician factors on achievable data rate and investigates the user scheduling strategy, combining scheme, power control, and dynamic bandwidth allocation to maximize the sum data rate in the distributed multiple-input-multiple-output (MIMO)-enabled low earth orbit (LEO) satellite networks. However, due to the resource-assignment problem, it is challenging to find the optimal solution for maximizing the sum data rate. To transform this problem into a more tractable form, we first quantify the channel estimation errors based on the minimum mean square error (MMSE) estimator and rigorously derive a closed-form lower bound of the achievable data rate, offering an explicit formulation for resource allocation. Then, to solve the NP-hard problem, we decompose it into three sub-problems, namely, user scheduling strategy, joint combination and power control, and dynamic bandwidth allocation, by using alternative optimization (AO). Specifically, the user scheduling is formulated as a graph coloring problem by iteratively updating an undirected graph based on user requirements, which is then solved using the DSatur algorithm. For the combining weights and power control, the successive convex approximation (SCA) and geometrical programming (GP) are adopted to obtain the sub-optimal solution with lower complexity. Finally, the optimal bandwidth allocation can be achieved by solving the concave problem. Numerical results validate the analytical tightness of the derived bound, especially for large Rician factors, and demonstrate significant performance gains over other benchmarks.

eess.SP↗

arXiv2025

SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training

Supervised fine-tuning (SFT) and reinforcement learning (RL) are widely used post-training techniques for foundation models. However, their roles in enhancing model generalization capabilities remain unclear. This paper studies the difference between SFT and RL on generalization and memorization, focusing on text-based rule variants and visual variants. We introduce GeneralPoints, an arithmetic reasoning card game, and adopt V-IRL, a real-world navigation environment, to assess how models trained with SFT and RL generalize to unseen variants in both textual and visual domains. We show that RL, especially when trained with an outcome-based reward, generalizes across both rule-based textual and visual variants. SFT, in contrast, tends to memorize training data and struggles to generalize out-of-distribution scenarios. Further analysis reveals that RL improves the model's underlying visual recognition capabilities, contributing to its enhanced generalization in the visual domain. Despite RL's superior generalization, we show that SFT remains essential for effective RL training; SFT stabilizes the model's output format, enabling subsequent RL to achieve its performance gains. These findings demonstrates the capability of RL for acquiring generalizable knowledge in complex, multi-modal tasks.

cs.AI↗