Chang Shu

Chang Shu, Jiuzhou Han, Fangyu Liu et al. · deepmind

Embodied language comprehension emphasizes that language understanding is not solely a matter of mental processing in the brain but also involves interactions with the physical and social environment. With the explosive growth of Large Language Models (LLMs) and their already ubiquitous presence in our daily lives, it is becoming increasingly necessary to verify their real-world understanding. Inspired by cognitive theories, we propose POSQA: a Physical Object Size Question Answering dataset with simple size comparison questions to examine the extremity and analyze the potential mechanisms of the embodied comprehension of the latest LLMs. We show that even the largest LLMs today perform poorly under the zero-shot setting. We then push their limits with advanced prompting techniques and external knowledge augmentation. Furthermore, we investigate whether their real-world comprehension primarily derives from contextual information or internal weights and analyse the impact of prompt formats and report bias of different objects. Our results show that real-world understanding that LLMs shaped from textual data can be vulnerable to deception and confusion by the surface form of prompts, which makes it less aligned with human behaviours.

Baian Chen, Chang Shu, Ehsan Shareghi et al.

Recent efforts have augmented language models (LMs) with external tools or environments, leading to the development of language agents that can reason and act. However, most of these agents rely on few-shot prompting techniques with off-the-shelf LMs. In this paper, we investigate and argue for the overlooked direction of fine-tuning LMs to obtain language agents. Using a setup of question answering (QA) with a Google search API, we explore a variety of base LMs, prompting methods, fine-tuning data, and QA tasks, and find language agents are consistently improved after fine-tuning their backbone LMs. For example, fine-tuning Llama2-7B with 500 agent trajectories generated by GPT-4 leads to a 77% HotpotQA performance increase. Furthermore, we propose FireAct, a novel approach to fine-tuning LMs with trajectories from multiple tasks and prompting methods, and show having more diverse fine-tuning data can further improve agents. Along with other findings regarding scaling effects, robustness, generalization, efficiency and cost, our work establishes comprehensive benefits of fine-tuning LMs for agents, and provides an initial set of experimental designs, insights, as well as open questions toward language agent fine-tuning.

Kaixin Zhang, Hongzhi Wang, Yabin Lu et al.

Learned cardinality estimation methods have achieved high precision compared to traditional methods. Among learned methods, query-driven approaches have faced the workload drift problem for a long time. Although both data-driven and hybrid methods are proposed to avoid this problem, most of them suffer from high training and estimation costs, limited scalability, instability, and long-tail distribution problems on high-dimensional tables, which seriously affects the practical application of learned cardinality estimators. In this paper, we prove that most of these problems are directly caused by the widely used progressive sampling. We solve this problem by introducing predicate information into the autoregressive model and propose Duet, a stable, efficient, and scalable hybrid method to estimate cardinality directly without sampling or any non-differentiable process, which can not only reduce the inference complexity from $O(n)$ to $O(1)$ compared to Naru and UAE but also achieve higher accuracy on high cardinality and high-dimensional tables. Experimental results show that Duet can achieve all the design goals above and be much more practical. Besides, Duet even has a lower inference cost on CPU than that of most learned methods on GPU.

Caiqi Zhang, Chang Shu, Ehsan Shareghi et al. · cambridge

Confidence estimation is essential for the reliable deployment of large language models (LLMs). Existing methods are primarily designed for factual QA tasks and often fail to generalize to reasoning tasks. To address this gap, we propose a set of training-free, graph-based confidence estimation methods tailored to reasoning tasks. Our approach models reasoning paths as directed graphs and estimates confidence by exploiting graph properties such as centrality, path convergence, and path weighting. Experiments with two LLMs on three reasoning datasets demonstrate improved confidence estimation and enhanced performance on two downstream tasks.

Yuxuan Mu, Hung Yu Ling, Yi Shi et al.

Motion capture (mocap) data often exhibits visually jarring artifacts due to inaccurate sensors and post-processing. Cleaning this corrupted data can require substantial manual effort from human experts, which can be a costly and time-consuming process. Previous data-driven motion cleanup methods offer the promise of automating this cleanup process, but often require in-domain paired corrupted-to-clean training data. Constructing such paired datasets requires access to high-quality, relatively artifact-free motion clips, which often necessitates laborious manual cleanup. In this work, we present StableMotion, a simple yet effective method for training motion cleanup models directly from unpaired corrupted datasets that need cleanup. The core component of our method is the introduction of motion quality indicators, which can be easily annotated - through manual labeling or heuristic algorithms - and enable training of quality-aware motion generation models on raw motion data with mixed quality. At test time, the model can be prompted to generate high-quality motions using the quality indicators. Our method can be implemented through a simple diffusion-based framework, leading to a unified motion generate-discriminate model, which can be used to both identify and fix corrupted frames. We demonstrate that our proposed method is effective for training motion cleanup models on raw mocap data in production scenarios by applying StableMotion to SoccerMocap, a 245-hour soccer mocap dataset containing real-world motion artifacts. The trained model effectively corrects a wide range of motion artifacts, reducing motion pops and frozen frames by 68% and 81%, respectively. Results and code are available at https://yxmu.foo/stablemotion-page

Joshua Li, Brendan Chharawala, Chang Shu et al.

Animating realistic character interactions with the surrounding environment is important for autonomous agents in gaming, AR/VR, and robotics. However, current methods for human motion reconstruction struggle with accurately placing humans in 3D space. We introduce Scene-Human Aligned REconstruction (SHARE), a technique that leverages the scene geometry's inherent spatial cues to accurately ground human motion reconstruction. Each reconstruction relies solely on a monocular RGB video from a stationary camera. SHARE first estimates a human mesh and segmentation mask for every frame, alongside a scene point map at keyframes. It iteratively refines the human's positions at these keyframes by comparing the human mesh against the human point map extracted from the scene using the mask. Crucially, we also ensure that non-keyframe human meshes remain consistent by preserving their relative root joint positions to keyframe root joints during optimization. Our approach enables more accurate 3D human placement while reconstructing the surrounding scene, facilitating use cases on both curated datasets and in-the-wild web videos. Extensive experiments demonstrate that SHARE outperforms existing methods.

Pengcheng Xi, Chang Shu, Rafik Goubran

Breast cancer is the most common cancer in women worldwide. The most common screening technology is mammography. To reduce the cost and workload of radiologists, we propose a computer aided detection approach for classifying and localizing calcifications and masses in mammogram images. To improve on conventional approaches, we apply deep convolutional neural networks (CNN) for automatic feature learning and classifier building. In computer-aided mammography, deep CNN classifiers cannot be trained directly on full mammogram images because of the loss of image details from resizing at input layers. Instead, our classifiers are trained on labelled image patches and then adapted to work on full mammogram images for localizing the abnormalities. State-of-the-art deep convolutional neural networks are compared on their performance of classifying the abnormalities. Experimental results indicate that VGGNet receives the best overall accuracy at 92.53\% in classifications. For localizing abnormalities, ResNet is selected for computing class activation maps because it is ready to be deployed without structural change or further training. Our approach demonstrates that deep convolutional neural network classifiers have remarkable localization capabilities despite no supervision on the location of abnormalities is provided.

Stefanie Wuhrer, Leonid Pishchulin, Alan Brunton et al.

Estimating the body shape and posture of a dressed human subject in motion represented as a sequence of (possibly incomplete) 3D meshes is important for virtual change rooms and security. To solve this problem, statistical shape spaces encoding human body shape and posture variations are commonly used to constrain the search space for the shape estimate. In this work, we propose a novel method that uses a posture-invariant shape space to model body shape variation combined with a skeleton-based deformation to model posture variation. Our method can estimate the body shape and posture of both static scans and motion sequences of dressed human body scans. In case of motion sequences, our method takes advantage of motion cues to solve for a single body shape estimate along with a sequence of posture estimates. We apply our approach to both static scans and motion sequences and demonstrate that using our method, higher fitting accuracy is achieved than when using a variant of the popular SCAPE model as statistical model.

Stefanie Wuhrer, Jochen Lang, Motahareh Tekieh et al.

We present an approach to robustly track the geometry of an object that deforms over time from a set of input point clouds captured from a single viewpoint. The deformations we consider are caused by applying forces to known locations on the object's surface. Our method combines the use of prior information on the geometry of the object modeled by a smooth template and the use of a linear finite element method to predict the deformation. This allows the accurate reconstruction of both the observed and the unobserved sides of the object. We present tracking results for noisy low-quality point clouds acquired by either a stereo camera or a depth camera, and simulations with point clouds corrupted by different error terms. We show that our method is also applicable to large non-linear deformations.

Augusto Salazar, Stefanie Wuhrer, Chang Shu et al.

We consider the problem of computing accurate point-to-point correspondences among a set of human face scans with varying expressions. Our fully automatic approach does not require any manually placed markers on the scan. Instead, the approach learns the locations of a set of landmarks present in a database and uses this knowledge to automatically predict the locations of these landmarks on a newly available scan. The predicted landmarks are then used to compute point-to-point correspondences between a template model and the newly available scan. To accurately fit the expression of the template to the expression of the scan, we use as template a blendshape model. Our algorithm was tested on a database of human faces of different ethnic groups with strongly varying expressions. Experimental results show that the obtained point-to-point correspondence is both highly accurate and consistent for most of the tested 3D face models.