深度学习的快速进步正在导致一系列快速变化的模型，对计算的需求急剧增长。但是，随着框架将性能优化专门针对流行网络的模式，它们隐含地限制了推动研究进展的新颖和多样化的模型。我们通过定义灵活和用户可定制的管道来优化基于数据运动最小化的任意深神经网络的培训来赋予深度学习研究人员的能力。管道始于Pytorch或ONNX中的标准网络，并通过逐步降低转换计算。我们定义了四个级别的通用转换级别，从局部操作员优化到全球数据运动减少。这些在以数据为中心的图形中间表示上运行，该表示在各个级别的抽象级别表达计算和数据移动，包括扩展基本运算符，例如其基础计算的卷积。设计的核心是管道的互动性和内省性质。每个部分都可以通过Python API扩展，并且可以使用GUI进行交互调整。我们在十个不同的网络上展示了竞争性能或加速性，交互式优化发现了高效网络中的新机会。

变形金刚是今天最重要的机器学习工作负载之一。培训是一个非常计算密集的任务，通常需要几天或几周，并且对优化变压器进行了重大关注。尽管如此，现有的实现不会有效地利用GPU。我们发现数据移动是培训时的关键瓶颈。由于Amdahl的法律和大规模改进的计算性能，培训现已成为记忆束缚。此外，现有框架使用次优数据布局。使用这些洞察力，我们提供了一个用于全局优化变压器数据移动的配方。我们将数据移动降低到22.91％，总体上实现了在训练伯特编码器层和1.19x的整个伯特的最先进框架上的1.30倍的性能改进。我们的方法更广泛地适用于优化深神经网络，并深入了解如何解决新兴的性能瓶颈。

In this paper, we perform an exhaustive evaluation of different representations to address the intent classification problem in a Spoken Language Understanding (SLU) setup. We benchmark three types of systems to perform the SLU intent detection task: 1) text-based, 2) lattice-based, and a novel 3) multimodal approach. Our work provides a comprehensive analysis of what could be the achievable performance of different state-of-the-art SLU systems under different circumstances, e.g., automatically- vs. manually-generated transcripts. We evaluate the systems on the publicly available SLURP spoken language resource corpus. Our results indicate that using richer forms of Automatic Speech Recognition (ASR) outputs allows SLU systems to improve in comparison to the 1-best setup (4% relative improvement). However, crossmodal approaches, i.e., learning from acoustic and text embeddings, obtains performance similar to the oracle setup, and a relative improvement of 18% over the 1-best configuration. Thus, crossmodal architectures represent a good alternative to overcome the limitations of working purely automatically generated textual data.

Autonomous driving is an exciting new industry, posing important research questions. Within the perception module, 3D human pose estimation is an emerging technology, which can enable the autonomous vehicle to perceive and understand the subtle and complex behaviors of pedestrians. While hardware systems and sensors have dramatically improved over the decades -- with cars potentially boasting complex LiDAR and vision systems and with a growing expansion of the available body of dedicated datasets for this newly available information -- not much work has been done to harness these novel signals for the core problem of 3D human pose estimation. Our method, which we coin HUM3DIL (HUMan 3D from Images and LiDAR), efficiently makes use of these complementary signals, in a semi-supervised fashion and outperforms existing methods with a large margin. It is a fast and compact model for onboard deployment. Specifically, we embed LiDAR points into pixel-aligned multi-modal features, which we pass through a sequence of Transformer refinement stages. Quantitative experiments on the Waymo Open Dataset support these claims, where we achieve state-of-the-art results on the task of 3D pose estimation.

We introduce Structured 3D Features, a model based on a novel implicit 3D representation that pools pixel-aligned image features onto dense 3D points sampled from a parametric, statistical human mesh surface. The 3D points have associated semantics and can move freely in 3D space. This allows for optimal coverage of the person of interest, beyond just the body shape, which in turn, additionally helps modeling accessories, hair, and loose clothing. Owing to this, we present a complete 3D transformer-based attention framework which, given a single image of a person in an unconstrained pose, generates an animatable 3D reconstruction with albedo and illumination decomposition, as a result of a single end-to-end model, trained semi-supervised, and with no additional postprocessing. We show that our S3F model surpasses the previous state-of-the-art on various tasks, including monocular 3D reconstruction, as well as albedo and shading estimation. Moreover, we show that the proposed methodology allows novel view synthesis, relighting, and re-posing the reconstruction, and can naturally be extended to handle multiple input images (e.g. different views of a person, or the same view, in different poses, in video). Finally, we demonstrate the editing capabilities of our model for 3D virtual try-on applications.

Domain adaptation has been vastly investigated in computer vision but still requires access to target images at train time, which might be intractable in some conditions, especially for long-tail samples. In this paper, we propose the task of `Prompt-driven Zero-shot Domain Adaptation', where we adapt a model trained on a source domain using only a general textual description of the target domain, i.e., a prompt. First, we leverage a pretrained contrastive vision-language model (CLIP) to optimize affine transformations of source features, bringing them closer to target text embeddings, while preserving their content and semantics. Second, we show that augmented features can be used to perform zero-shot domain adaptation for semantic segmentation. Experiments demonstrate that our method significantly outperforms CLIP-based style transfer baselines on several datasets for the downstream task at hand. Our prompt-driven approach even outperforms one-shot unsupervised domain adaptation on some datasets, and gives comparable results on others. The code is available at https://github.com/astra-vision/PODA.

Graph learning problems are typically approached by focusing on learning the topology of a single graph when signals from all nodes are available. However, many contemporary setups involve multiple related networks and, moreover, it is often the case that only a subset of nodes is observed while the rest remain hidden. Motivated by this, we propose a joint graph learning method that takes into account the presence of hidden (latent) variables. Intuitively, the presence of the hidden nodes renders the inference task ill-posed and challenging to solve, so we overcome this detrimental influence by harnessing the similarity of the estimated graphs. To that end, we assume that the observed signals are drawn from a Gaussian Markov random field with latent variables and we carefully model the graph similarity among hidden (latent) nodes. Then, we exploit the structure resulting from the previous considerations to propose a convex optimization problem that solves the joint graph learning task by providing a regularized maximum likelihood estimator. Finally, we compare the proposed algorithm with different baselines and evaluate its performance over synthetic and real-world graphs.

Integration of multiple sensor modalities and deep learning into Simultaneous Localization And Mapping (SLAM) systems are areas of significant interest in current research. Multi-modality is a stepping stone towards achieving robustness in challenging environments and interoperability of heterogeneous multi-robot systems with varying sensor setups. With maplab 2.0, we provide a versatile open-source platform that facilitates developing, testing, and integrating new modules and features into a fully-fledged SLAM system. Through extensive experiments, we show that maplab 2.0's accuracy is comparable to the state-of-the-art on the HILTI 2021 benchmark. Additionally, we showcase the flexibility of our system with three use cases: i) large-scale (approx. 10 km) multi-robot multi-session (23 missions) mapping, ii) integration of non-visual landmarks, and iii) incorporating a semantic object-based loop closure module into the mapping framework. The code is available open-source at https://github.com/ethz-asl/maplab.

When developing deep learning models, we usually decide what task we want to solve then search for a model that generalizes well on the task. An intriguing question would be: what if, instead of fixing the task and searching in the model space, we fix the model and search in the task space? Can we find tasks that the model generalizes on? How do they look, or do they indicate anything? These are the questions we address in this paper. We propose a task discovery framework that automatically finds examples of such tasks via optimizing a generalization-based quantity called agreement score. We demonstrate that one set of images can give rise to many tasks on which neural networks generalize well. These tasks are a reflection of the inductive biases of the learning framework and the statistical patterns present in the data, thus they can make a useful tool for analysing the neural networks and their biases. As an example, we show that the discovered tasks can be used to automatically create adversarial train-test splits which make a model fail at test time, without changing the pixels or labels, but by only selecting how the datapoints should be split between the train and test sets. We end with a discussion on human-interpretability of the discovered tasks.

Tabular biomedical data is often high-dimensional but with a very small number of samples. Although recent work showed that well-regularised simple neural networks could outperform more sophisticated architectures on tabular data, they are still prone to overfitting on tiny datasets with many potentially irrelevant features. To combat these issues, we propose Weight Predictor Network with Feature Selection (WPFS) for learning neural networks from high-dimensional and small sample data by reducing the number of learnable parameters and simultaneously performing feature selection. In addition to the classification network, WPFS uses two small auxiliary networks that together output the weights of the first layer of the classification model. We evaluate on nine real-world biomedical datasets and demonstrate that WPFS outperforms other standard as well as more recent methods typically applied to tabular data. Furthermore, we investigate the proposed feature selection mechanism and show that it improves performance while providing useful insights into the learning task.