arXiv subjects

Stephen Wang

Publications and source records attributed to Stephen Wang.

14 recordsLinked to original sources

arXiv2026

Forecast Collapse in Time-Series Foundation Models

When forecasting hourly returns for 1,000 US equities, we observe an unexpected phenomenon: predictions become nearly flat and show poor stock ranking, as measured by cross-sectional correlation. We call this forecast collapse. Surprisingly, the phenomenon largely disappears when forecasting trading volume under the same setting. We investigate forecast collapse across time-series foundation models (TSFMs), twelve deep-learning forecasting models, and 97 public benchmark configurations, and find that it is closely tied to target predictability. We identify two distinct reasons behind it: low predictability limits the amplitude of calibrated point forecasts, while per-series objectives leave cross-series structure unidentified. These findings reveal a calibration-ranking tradeoff: optimizing squared error leads to flat predictions, whereas directly optimizing cross-sectional correlation improves ranking but can inflate forecast amplitude by more than an order of magnitude. To address this tradeoff, we introduce CalibRank, a simple objective that balances calibration and ranking. On Finance1K, CalibRank nearly triples cross-sectional correlation while keeping amplitude close to the target, and improves correlation on all tested models. Our results reveal a blind spot in conventional time-series evaluation: per-series metrics can hide failures in cross-series structure needed by downstream decisions.

cs.LG↗

arXiv2026

Hybrid Self-evolving Structured Memory for GUI Agents

The remarkable progress of vision-language models (VLMs) has enabled GUI agents to interact with computers in a human-like manner. Yet real-world computer-use tasks remain difficult due to long-horizon workflows, diverse interfaces, and frequent intermediate errors. Prior work equips agents with external memory built from large collections of trajectories, but relies on flat retrieval over discrete summaries or continuous embeddings, falling short of the structured organization and self-evolving characteristics of human memory. Inspired by the brain, we propose Hybrid Self-evolving Structured Memory (HyMEM), a graph-based memory that couples discrete high-level symbolic nodes with continuous trajectory embeddings. HyMEM maintains a graph structure to support multi-hop retrieval, self-evolution via node update operations, and on-the-fly working-memory refreshing during inference. Extensive experiments show that HyMEM consistently improves open-source GUI agents, enabling 7B/8B backbones to match or surpass strong closed-source models; notably, it boosts Qwen2.5-VL-7B by +22.5% and outperforms Gemini2.5-Pro-Vision and GPT-4o.

cs.AI↗

arXiv2026

Transformer Is Inherently a Causal Learner

We reveal that transformers trained in an autoregressive manner naturally encode time-delayed causal structures in their learned representations. When predicting future values in multivariate time series, the gradient sensitivities of transformer outputs with respect to past inputs directly recover the underlying causal graph, without any explicit causal objectives or structural constraints. We prove this connection theoretically under standard identifiability conditions and develop a practical extraction method using aggregated gradient attributions. On challenging cases such as nonlinear dynamics, long-term dependencies, and non-stationary systems, this approach greatly surpasses the performance of state-of-the-art discovery algorithms, especially as data heterogeneity increases, exhibiting scaling potential where causal accuracy improves with data volume and heterogeneity, a property traditional methods lack. This unifying view lays the groundwork for a future paradigm where causal discovery operates through the lens of foundation models, and foundation models gain interpretability and enhancement through the lens of causality.

cs.LG↗

arXiv2025

NeuromorphicRx: From Neural to Spiking Receiver

In this work, we propose a novel energy-efficient spiking neural network (SNN)-based receiver for 5G-NR OFDM system, called neuromorphic receiver (NeuromorphicRx), replacing the channel estimation, equalization and symbol demapping blocks. We leverage domain knowledge to design the input with spiking encoding and propose a deep convolutional SNN with spike-element-wise residual connections. We integrate an SNN with artificial neural network (ANN) hybrid architecture to obtain soft outputs and employ surrogate gradient descent for training. We focus on generalization across diverse scenarios and robustness through quantized aware training. We focus on interpretability of NeuromorphicRx for 5G-NR signals and perform detailed ablation study for 5G-NR signals. Our extensive numerical simulations show that NeuromorphicRx is capable of achieving significant block error rate performance gain compared to 5G-NR receivers and similar performance compared to its ANN-based counterparts with 7.6x less energy consumption.

cs.NE↗

arXiv2025

Low-Complexity OFDM Deep Neural Receivers

Deep neural receivers (NeuralRxs) for Orthogonal Frequency Division Multiplexing (OFDM) signals are proposed for enhanced decoding performance compared to their signal-processing based counterparts. However, the existing architectures ignore the required number of epochs for training convergence and floating-point operations (FLOPs), which increase significantly with improving performance. To tackle these challenges, we propose a new residual network (ResNet) block design for OFDM NeuralRx. Specifically, we leverage small kernel sizes and dilation rates to lower the number of FLOPs (NFLOPs) and uniform channel sizes to reduce the memory access cost (MAC). The ResNet block is designed with novel channel split and shuffle blocks, element-wise additions are removed, with Gaussian error linear unit (GELU) activations. Extensive simulations show that our proposed NeuralRx reduces NFLOPs and improves training convergence while improving the decoding accuracy.

cs.IT↗

arXiv2025

Auto-scaling Continuous Memory for GUI Agent

We study how to endow GUI agents with scalable memory that help generalize across unfamiliar interfaces and long-horizon tasks. Prior GUI agents compress past trajectories into text tokens, which balloons context length and misses decisive visual cues (e.g., exact widget size and position). We propose a continuous memory that encodes each GUI trajectory into a fixed-length sequence of continuous embeddings using the VLM itself as an encoder; these embeddings are plugged directly into the backbone's input layer, sharply reducing context cost while preserving fine-grained visual information. As memory size and retrieval depth increase, performance improves monotonically, unlike text memories that degrade with long prompts. To grow memory at low cost, we introduce an auto-scaling data flywheel that (i) discovers new environments via search, (ii) synthesizes tasks with an open-source VLM, (iii) rolls out trajectories with the agent, and (iv) verifies success with the same VLM. Using this pipeline, we collect 100k+ trajectories for about \$4000 and fine-tune only the memory encoder (LoRA on a Q-Former, 1.2\% parameters) with 1,500 samples. On real-world GUI benchmarks, our memory-augmented agent consistently improves success rates under long horizons and distribution shifts. Notably, Qwen-2.5-VL-7B + continuous memory achieves performance comparable to state-of-the-art closed-source models (e.g., GPT-4o, Claude-4).

cs.AI↗

arXiv2025

Backdoor Attribution: Elucidating and Controlling Backdoor in Language Models

Fine-tuned Large Language Models (LLMs) are vulnerable to backdoor attacks through data poisoning, yet the internal mechanisms governing these attacks remain a black box. Previous research on interpretability for LLM safety tends to focus on alignment, jailbreak, and hallucination, but overlooks backdoor mechanisms, making it difficult to understand and fully eliminate the backdoor threat. In this paper, aiming to bridge this gap, we explore the interpretable mechanisms of LLM backdoors through Backdoor Attribution (BkdAttr), a tripartite causal analysis framework. We first introduce the Backdoor Probe that proves the existence of learnable backdoor features encoded within the representations. Building on this insight, we further develop Backdoor Attention Head Attribution (BAHA), efficiently pinpointing the specific attention heads responsible for processing these features. Our primary experiments reveals these heads are relatively sparse; ablating a minimal \textbf{$\sim$ 3%} of total heads is sufficient to reduce the Attack Success Rate (ASR) by \textbf{over 90%}. More importantly, we further employ these findings to construct the Backdoor Vector derived from these attributed heads as a master controller for the backdoor. Through only \textbf{1-point} intervention on \textbf{single} representation, the vector can either boost ASR up to \textbf{$\sim$ 100% ($\uparrow$)} on clean inputs, or completely neutralize backdoor, suppressing ASR down to \textbf{$\sim$ 0% ($\downarrow$)} on triggered inputs. In conclusion, our work pioneers the exploration of mechanistic interpretability in LLM backdoors, demonstrating a powerful method for backdoor control and revealing actionable insights for the community.

cs.CR↗

arXiv2025

Rate-Splitting Multiple Access for 6G: Prototypes, Experimental Results and Link/System level Simulations

Rate-Splitting Multiple Access (RSMA) is a powerful and versatile physical layer multiple access technique that generalizes and has better interference management capabilities than 5G-based Space Division Multiple Access (SDMA). It is also a rapidly maturing technology, all of which makes it a natural successor to SDMA in 6G. In this article, we describe RSMA's suitability for 6G by presenting: i) link and system level simulations of RSMA's performance gains over SDMA in realistic environments, and (ii) pioneering experimental results that demonstrate RSMA's gains over SDMA for key use cases like enhanced Mobile Broadband (eMBb), and Integrated Sensing and Communications (ISAC). We also comment on the status of standardization activities for RSMA.

eess.SP↗

arXiv2024

SpikingRx: From Neural to Spiking Receiver

In this work, we propose an energy efficient neuromorphic receiver to replace multiple signal-processing blocks at the receiver by a Spiking Neural Network (SNN) based module, called SpikingRx. We propose a deep convolutional SNN with spike-element-wise ResNet layers which takes a whole OFDM grid compliant with 5G specifications and provides soft outputs for decoded bits that can be used as log-likelihood ratios. We propose to employ the surrogate gradient descent method for training the SpikingRx and focus on its generalizability and robustness to quantization. Moreover, the interpretability of the proposed SpikingRx is studied by a comprehensive ablation study. Our extensive numerical simulations show that SpikingRx is capable of achieving significant block error rate performance gain compared to conventional 5G receivers and similar performance compared to its traditional NN-based counterparts with approximately 9x less energy consumption.

cs.IT↗

arXiv2023

Rate-Splitting Multiple Access for Semantic-Aware Networks: an Age of Incorrect Information Perspective

In this letter, we design a downlink multi-user communication framework based on Rate-Splitting Multiple Access (RSMA) for semantic-aware networks. First, we formulate an optimization problem to obtain the optimal user scheduling, precoding, and power allocation schemes jointly. We consider the metric Age of Incorrect Information (AoII) in the objective function of the formulated problem to maximize the freshness of the overall information to be transmitted. Using big-M and Successive Convex Approximation (SCA) methods, we convert the resulting non-convex problem with conditional objective and constraints into a convex one and propose an iterative algorithm to solve it. By numerical results, we show that RSMA achieves a lower AoII than SDMA owing to its superior performance under multi-user interference.

cs.IT↗

arXiv2023

A Proof of Concept for OTFS Resilience in Doubly-Selective Channels by GPU-Enabled Real-Time SDR

Orthogonal time frequency space (OTFS) is a modulation technique which is robust against the disruptive effects of doubly-selective channels. In this paper, we perform an experimental study of OTFS by a real-time software defined radio (SDR) setup. Our SDR consists of a Graphical Processing Unit (GPU) for signal processing programmed using Sionna and TensorFlow, and Universal Software Radio Peripheral (USRP) devices for air interface. We implement a low-latency transceiver structure for OTFS and investigate its performance under various Doppler values. By comparing the performance of OTFS with Orthogonal Frequency Division Multiplexing (OFDM), we demonstrate that OTFS is highly robust against the disruptive effects of doubly-selective channels in a real-time experimental setup.

cs.IT↗

arXiv2023

A Low-Complexity Design for Rate-Splitting Multiple Access in Overloaded MIMO Networks

Rate-Splitting Multiple Access (RSMA) is a robust multiple access scheme for multi-antenna wireless networks. In this work, we study the performance of RSMA in downlink overloaded networks, where the number of transmit antennas is smaller than the number of users. We provide analysis and closed-form solutions for optimal power and rate allocations that maximize max-min fairness when low-complexity precoding schemes are employed. The derived closed-form solutions are used to propose a low-complexity RSMA system design for precoder selection and resource allocation for arbitrary number of users and antennas under perfect Channel State Information at the Transmitter (CSIT). We compare the performance of the proposed design with benchmark designs based on Space Division Multiple Access (SDMA) to show that the proposed low-complexity RSMA design achieves a significantly higher performance gain in overloaded networks.

cs.IT↗

arXiv2023

RSMA for Overloaded MIMO Networks: Low-Complexity Design for Max-Min Fairness

Rate-Splitting Multiple Access (RSMA) is a robust multiple access scheme for multi-antenna wireless networks. In this work, we study the performance of RSMA in downlink overloaded networks, where the number of transmit antennas is smaller than the number of users. We provide analysis and closed-form solutions for optimal power and rate allocations that maximize max-min fairness when low-complexity precoding schemes are employed. The derived closed-form solutions are used to propose a low-complexity RSMA system design for precoder selection and resource allocation for arbitrary number of users and antennas under perfect and imperfect Channel State Information at the Transmitter (CSIT). We compare the performance of the proposed design with benchmark designs based on Space Division Multiple Access (SDMA) with and without user scheduling. By numerical results, we show that the proposed low-complexity RSMA design achieves a significantly higher rate compared to the SDMA-based benchmark designs under perfect and imperfect CSIT.

cs.IT↗

arXiv2007

Representations of Surface Groups and Right-Angled Artin Groups in Higher Rank

We give very flexible, concrete constructions of discrete and faithful epresentations of right-angled Artin groups into higher-rank Lie groups. Using the geometry of the associated symmetric spaces and the combinatorics of the groups, we find a general criterion for when discrete and faithful representations exist, and show that the criterion is satisfied in particular cases. There are direct applications towards constructing representations of surface groups into higher-rank Lie groups, and, in particular, into lattices in higher-rank Lie groups.

math.GR↗