Long-Giang Vu

Denoising Probabilistic Models (DPMs) represent an emerging domain of generative models that excel in generating di- verse and high-quality images. However, most current training methods for DPMs often neglect the correlation between timesteps, limiting the model’s performance in generating images effectively. Notably, we theoretically point out that this issue can be caused by the cumulative estimation gap between the predicted and the actual trajectory. To minimize that gap, we propose a novel sequence-aware loss that aims to reduce the estimation gap to enhance the sampling quality. Furthermore, we theoretically show that our proposed loss function is a tighter upper bound of the estimation loss in comparison with the conventional loss in DPMs. Experimental results on several benchmark datasets including CIFAR10, CelebA, and CelebA-HQ consistently show a remarkable improvement of our proposed method regarding the image generalization quality measured by FID and Inception Score compared to several DPM baselines. Our code and pre-trained checkpoints are available at https://github.com/VinAIResearch/SA-DPM.

Robustness and generalization ability of machine learning models are of utmost importance in various application domains. There is a wide interest in efficient ways to analyze those properties. One important direction is to analyze connection between those two properties. Prior theories suggest that a robust learning algorithm can produce trained models with a high generalization ability. However, we show in this work that the existing error bounds are vacuous for the Bayes optimal classifier which is the best among all measurable classifiers for a classification problem with overlapping classes. Those bounds cannot converge to the true error of this ideal classifier. This is undesirable, surprizing, and never known before. We then present a class of novel bounds, which are model-dependent and provably tighter than the existing robustness-based ones. Unlike prior ones, our bounds are guaranteed to converge to the true error of the best classifier, as the number of samples increases. We further provide an extensive experiment and find that two of our bounds are often non-vacuous for a large class of deep neural networks, pretrained from ImageNet.

Vision-Language Navigation (VLN) in previously unseen indoor environments is useful in real-world robotics, where an agent must follow natural-language instructions, locate referred objects, and answer spatial questions without a pre-built map or a fixed object vocabulary. Recently, multimodal vision- language models (VLMs) have shown strong open-vocabulary grounding and zero-shot reasoning, making them a natural semantic front-end for such agents. One common problem is these VLMs’ inability to emit reliable metric quantities such as range, bearing, and comparative spatial relations directly from images. Existing approaches address this by folding geometry into a hand-engineered pipeline, or by asking the model to output waypoints; both require rewriting the control stack for each new robot, task, or vocabulary, which does not scale. This work designs AnchorVLN, an open-vocabulary VLN system built on a single rule (the VLM proposes semantics; geometry decides metrics), realised as EMBODIED-NAV-MCP, a Model Context Protocol (MCP) server that a VLM agent drives at inference time through a compact set of callable tools. Because no tool accepts a distance in metres or a bearing in radians, the schema itself enforces the boundary between semantics and geometry without rewriting the downstream autonomy stack. We benchmark the server on both tasks of the CMU Vision-Language Navigation Challenge 2026: all 30 instruction-following questions over 15 scenes, and a frozen 45-question object-reference set on the same scenes. The full system achieves 64.4% on instruction following, dropping by 13.3 percentage points without controller modeling (t = 2.77, paired over questions). On object reference, geometric anchoring clears the challenge’s overlap threshold on 10 of 45 questions (versus 0 of 45 for direct model coordinate estimation), reducing median center error from 3.37 m to 2.48 m.