arXiv subjects

Haoyu Tan

Publications and source records attributed to Haoyu Tan.

3 recordsLinked to original sources

arXiv2026

Wuying-Browser-Agent: Real-World Centric Fundamental Long-Horizon Browser Agents

Browser agents perform well on short, clean demonstrations, but real deployment is fundamentally different: agents must sustain dozens of decisions on live websites while recovering from mistakes and navigating complex UIs. We argue that closing this gap requires alignment at every level of the pipeline, including execution, supervision, optimization, and evaluation, rather than scale alone. We present Wuying-Browser-Agent, a unified framework that addresses each of these levels. A structured browser harness provides stable execution primitives and decision-oriented context management. Reflection and UI-specialized Curriculum SFT (RUIC-SFT) explicitly trains on recovery trajectories and complex-UI interactions. Divergence-Aware Online GRPO (DAO-GRPO) improves long-horizon credit assignment through potential-based reward shaping and divergence-aware step weighting. Finally, we introduce BrowserBench, a bilingual real-web benchmark of 350 tasks averaging 37.9 steps, because most existing benchmarks are too short to expose long-horizon failure modes. Wuying-Browser-Agent-27B achieves 80.6\% on WebVoyager, 66.7\% on Online-Mind2Web, and 65.1\% on BrowserBench, establishing a new open-source state of the art on browser-use benchmarks. The same pipeline also transfers beyond browser use, demonstrating strong general agentic ability and reaching an average score of 73.8 on Tau2-Bench, Claw-Eval, and BFCL-v4.

cs.AI↗

arXiv2026

Efficient Sparse Selective-Update RNNs for Long-Range Sequence Modeling

Real-world sequential signals, such as audio or video, contain critical information that is often embedded within long periods of silence or noise. While recurrent neural networks (RNNs) are designed to process such data efficiently, they often suffer from ``memory decay'' due to a rigid update schedule: they typically update their internal state at every time step, even when the input is static. This constant activity forces the model to overwrite its own memory and makes it hard for the learning signal to reach back to distant past events. Here we show that we can overcome this limitation using Selective-Update RNNs (suRNNs), a non-linear architecture that learns to preserve its memory when the input is redundant. By using a neuron-level binary switch that only opens for informative events, suRNNs decouple the recurrent updates from the raw sequence length. This mechanism allows the model to maintain an exact, unchanged memory of the past during low-information intervals, creating a direct path for gradients to flow across time. Our experiments on the Long Range Arena, WikiText, and other synthetic benchmarks show that suRNNs match or exceed the accuracy of much more complex models such as Transformers, while remaining significantly more efficient for long-term storage. By allowing each neuron to learn its own update timescale, our approach resolves the mismatch between how long a sequence is and how much information it actually contains. By providing a principled approach to managing temporal information density, this work establishes a new direction for achieving Transformer-level performance within the highly efficient framework of recurrent modeling.

cs.LG↗

arXiv2025

Enhancing Foveated Rendering with Weighted Reservoir Sampling

Spatiotemporal sensitivity to high frequency information declines with increased peripheral eccentricity. Foveated rendering exploits this by decreasing the spatial resolution of rendered images in peripheral vision, reducing the rendering cost by omitting high frequency details. As foveation levels increase, the rendering quality is reduced, and traditional foveated rendering systems tend not to preserve samples that were previously rendered at high spatial resolution in previous frames. Additionally, prior research has shown that saccade landing positions are distributed around a target location rather than landing at a single point, and that even during fixations, eyes perform small microsaccades around a fixation point. This creates an opportunity for sampling from temporally neighbouring frames with differing foveal locations to reduce the required rendered size of the foveal region while achieving a higher perceived image quality. We further observe that the temporal presentation of pixels frame-to-frame can be viewed as a data stream, presenting a random sampling problem. Following this intuition, we propose a Weighted Reservoir Sampling technique to efficiently maintain a reservoir of the perceptually relevant high quality pixel samples from previous frames and incorporate them into the computation of the current frame. This allows the renderer to render a smaller region of foveal pixels per frame by temporally reusing pixel samples that are still relevant to reconstruct a higher perceived image quality, while allowing for higher levels of foveation. Our method operates on the output of foveated rendering, and runs in under 1\,ms at 4K resolution, making it highly efficient and integrable with real-time VR and AR foveated rendering systems.

cs.GR↗