Andrew Fitzgibbon

Alberto Cattaneo, Daniel Justus, Harry Mellor et al.

We present the award-winning submission to the WikiKG90Mv2 track of OGB-LSC@NeurIPS 2022. The task is link-prediction on the large-scale knowledge graph WikiKG90Mv2, consisting of 90M+ nodes and 600M+ edges. Our solution uses a diverse ensemble of $85$ Knowledge Graph Embedding models combining five different scoring functions (TransE, TransH, RotatE, DistMult, ComplEx) and two different loss functions (log-sigmoid, sampled softmax cross-entropy). Each individual model is trained in parallel on a Graphcore Bow Pod$_{16}$ using BESS (Balanced Entity Sampling and Sharing), a new distribution framework for KGE training and inference based on balanced collective communications between workers. Our final model achieves a validation MRR of 0.2922 and a test-challenge MRR of 0.2562, winning the first place in the competition. The code is publicly available at: https://github.com/graphcore/distributed-kge-poplar/tree/2022-ogb-submission.

Jan Svoboda, Thomas Cashman, Andrew Fitzgibbon

Embedded computer vision applications increasingly require the speed and power benefits of single-precision (32 bit) floating point. However, applications which make use of Levenberg-like optimization can lose significant accuracy when reducing to single precision, sometimes unrecoverably so. This accuracy can be regained using solvers based on QR rather than Cholesky decomposition, but the absence of sparse QR solvers for common sparsity patterns found in computer vision means that many applications cannot benefit. We introduce an open-source suite of solvers for Eigen, which efficiently compute the QR decomposition for matrices with some common sparsity patterns (block diagonal, horizontal and vertical concatenation, and banded). For problems with very particular sparsity structures, these elements can be composed together in 'kit' form, hence the name QRkit. We apply our methods to several computer vision problems, showing competitive performance and suitability especially in single precision arithmetic.

Sadegh Aliakbarian, Pashmina Cameron, Federica Bogo et al.

To represent people in mixed reality applications for collaboration and communication, we need to generate realistic and faithful avatar poses. However, the signal streams that can be applied for this task from head-mounted devices (HMDs) are typically limited to head pose and hand pose estimates. While these signals are valuable, they are an incomplete representation of the human body, making it challenging to generate a faithful full-body avatar. We address this challenge by developing a flow-based generative model of the 3D human body from sparse observations, wherein we learn not only a conditional distribution of 3D human pose, but also a probabilistic mapping from observations to the latent space from which we can generate a plausible pose along with uncertainty estimates for the joints. We show that our approach is not only a strong predictive model, but can also act as an efficient pose prior in different optimization settings where a good initial latent code plays a major role.

Andrew Fitzgibbon, Christoph M. Wintersteiger, Jeffrey Sarnoff

The IEEE P3109 draft standard defines a parameterized family of binary floating-point formats and associated operations, with a focus on facilitating machine learning. These formats allow efficient and consistent representation of values in a small number of bits. The defined formats are parameterized over width and precision in bits, signedness, and the presence of infinities. Operations are defined by decoding floating-point values to the set of closed extended reals: the reals augmented with positive and negative infinity and NaN (Not a Number). Explicit treatment of NaN and infinite operands ensures that only real arithmetic is invoked in operation definitions. Extensive rounding and saturation modes are defined; stochastic rounding is included. Operations are exception-free, accelerating throughput, with exceptional situations communicated through return values, e.g., NaN. Operations on blocks of values sharing a common scale factor are defined in terms of the underlying operations in a uniform manner. System vendors may describe approximate implementations via a novel scale-invariant measure, akin to units in the last place, called kappa-approximation. Standard function definitions and various other properties are mechanically verified and generated using formal specifications.

Dominic Masters, Josef Dean, Kerstin Klaser et al.

We present GPS++, a hybrid Message Passing Neural Network / Graph Transformer model for molecular property prediction. Our model integrates a well-tuned local message passing component and biased global attention with other key ideas from prior literature to achieve state-of-the-art results on large-scale molecular dataset PCQM4Mv2. Through a thorough ablation study we highlight the impact of individual components and find that nearly all of the model's performance can be maintained without any use of global self-attention, showing that message passing is still a competitive approach for 3D molecular property prediction despite the recent dominance of graph transformers. We also find that our approach is significantly more accurate than prior art when 3D positional information is not available.

Alexander Mathiasen, Hatem Helal, Kerstin Klaser et al.

The emergence of foundation models in Computer Vision and Natural Language Processing have resulted in immense progress on downstream tasks. This progress was enabled by datasets with billions of training examples. Similar benefits are yet to be unlocked for quantum chemistry, where the potential of deep learning is constrained by comparatively small datasets with 100k to 20M training examples. These datasets are limited in size because the labels are computed using the accurate (but computationally demanding) predictions of Density Functional Theory (DFT). Notably, prior DFT datasets were created using CPU supercomputers without leveraging hardware acceleration. In this paper, we take a first step towards utilising hardware accelerators by introducing the data generator PySCF$_{\text{IPU}}$ using Intelligence Processing Units (IPUs). This allowed us to create the dataset QM1B with one billion training examples containing 9-11 heavy atoms. We demonstrate that a simple baseline neural network (SchNet 9M) improves its performance by simply increasing the amount of training data without additional inductive biases. To encourage future researchers to use QM1B responsibly, we highlight several limitations of QM1B and emphasise the low-resolution of our DFT options, which also serves as motivation for even larger, more accurate datasets. Code and dataset are available on Github: http://github.com/graphcore-research/pyscf-ipu

Dominique Beaini, Shenyang Huang, Joao Alex Cunha et al.

Recently, pre-trained foundation models have enabled significant advancements in multiple fields. In molecular machine learning, however, where datasets are often hand-curated, and hence typically small, the lack of datasets with labeled features, and codebases to manage those datasets, has hindered the development of foundation models. In this work, we present seven novel datasets categorized by size into three distinct categories: ToyMix, LargeMix and UltraLarge. These datasets push the boundaries in both the scale and the diversity of supervised labels for molecular learning. They cover nearly 100 million molecules and over 3000 sparsely defined tasks, totaling more than 13 billion individual labels of both quantum and biological nature. In comparison, our datasets contain 300 times more data points than the widely used OGB-LSC PCQM4Mv2 dataset, and 13 times more than the quantum-only QM1B dataset. In addition, to support the development of foundational models based on our proposed datasets, we present the Graphium graph machine learning library which simplifies the process of building and training molecular machine learning models for multi-task and multi-level molecular datasets. Finally, we present a range of baseline results as a starting point of multi-task and multi-level training on these datasets. Empirically, we observe that performance on low-resource biological datasets show improvement by also training on large amounts of quantum data. This indicates that there may be potential in multi-task and multi-level training of a foundation model and fine-tuning it to resource-constrained downstream tasks.

Amir Shaikhha, Mathieu Huot, Shabnam Ghasemirad et al.

Automatic differentiation (AD) is a technique for computing the derivative of a function represented by a program. This technique is considered as the de-facto standard for computing the differentiation in many machine learning and optimisation software tools. Despite the practicality of this technique, the performance of the differentiated programs, especially for functional languages and in the presence of vectors, is suboptimal. We present an AD system for a higher-order functional array-processing language. The core functional language underlying this system simultaneously supports both source-to-source forward-mode AD and global optimisations such as loop transformations. In combination, gradient computation with forward-mode AD can be as efficient as reverse mode, and the Jacobian matrices required for numerical algorithms such as Gauss-Newton and Levenberg-Marquardt can be efficiently computed.

Paul Balança, Sam Hosegood, Carlo Luschi et al.

Low-precision formats such as float8 have been introduced in machine learning accelerated hardware to improve computational efficiency for large language models training and inference. Nevertheless, adoption by the ML community has been slowed down by the complex, and sometimes brittle, techniques required to match higher precision training accuracy. In this work, we present Scalify, a end-to-end scale propagation paradigm for computational graphs, generalizing and formalizing existing tensor scaling methods. Experiment results show that Scalify supports out-of-the-box float8 matrix multiplication and gradients representation, as well as float16 optimizer state storage. Our JAX implementation of Scalify is open-sourced at https://github.com/graphcore-research/jax-scalify

Kerstin Kläser, Błażej Banaszewski, Samuel Maddrell-Mander et al.

In biological tasks, data is rarely plentiful as it is generated from hard-to-gather measurements. Therefore, pre-training foundation models on large quantities of available data and then transfer to low-data downstream tasks is a promising direction. However, how to design effective foundation models for molecular learning remains an open question, with existing approaches typically focusing on models with large parameter capacities. In this work, we propose $\texttt{MiniMol}$, a foundational model for molecular learning with 10 million parameters. $\texttt{MiniMol}$ is pre-trained on a mix of roughly 3300 sparsely defined graph- and node-level tasks of both quantum and biological nature. The pre-training dataset includes approximately 6 million molecules and 500 million labels. To demonstrate the generalizability of $\texttt{MiniMol}$ across tasks, we evaluate it on downstream tasks from the Therapeutic Data Commons (TDC) ADMET group showing significant improvements over the prior state-of-the-art foundation model across 17 tasks. $\texttt{MiniMol}$ will be a public and open-sourced model for future research.

Andrew Fitzgibbon, Stephen Felix

Large-scale numerical computations make increasing use of low-precision (LP) floating point formats and mixed precision arithmetic, which can be enhanced by the technique of stochastic rounding (SR), that is, rounding an intermediate high-precision value up or down randomly as a function of the value's distance to the two rounding candidates. Stochastic rounding requires, in addition to the high-precision input value, a source of random bits. As the provision of high-quality random bits is an additional computational cost, it is of interest to require as few bits as possible while maintaining the desirable properties of SR in a given computation, or computational domain. This paper examines a number of possible implementations of few-bit stochastic rounding (FBSR), and shows how several natural implementations can introduce sometimes significant bias into the rounding process, which are not present in the case of infinite-bit, infinite-precision examinations of these implementations. The paper explores the impact of these biases in machine learning examples, and hence opens another class of configuration parameters of which practitioners should be aware when developing or adopting low-precision floating point. Code is available at http://github.com/graphcore-research/arith25-stochastic-rounding.

Hatem Helal, Andrew Fitzgibbon

Electronic structure simulation (ESS) has been used for decades to provide quantitative scientific insights on an atomistic scale, enabling advances in chemistry, biology, and materials science, among other disciplines. Following standard practice in scientific computing, the software packages driving these studies have been implemented in compiled languages such as FORTRAN and C. However, the recent introduction of machine learning (ML) into these domains has meant that ML models must be coded in these languages, or that complex software bridges have to be built between ML models in Python and these large compiled software systems. This is in contrast with recent progress in modern ML frameworks which aim to optimise both ease of use and high performance by harnessing hardware acceleration of tensor programs defined in Python. We introduce MESS: a modern electronic structure simulation package implemented in JAX; porting the ESS code to the ML world. We outline the costs and benefits of following the software development practices used in ML for this important scientific workload. MESS shows significant speedups n widely available hardware accelerators and simultaneously opens a clear pathway towards combining ESS with ML. MESS is available at https://github.com/graphcore-research/mess.

Filip Šrajer, Zuzana Kukelova, Andrew Fitzgibbon

Algorithmic differentiation (AD) allows exact computation of derivatives given only an implementation of an objective function. Although many AD tools are available, a proper and efficient implementation of AD methods is not straightforward. The existing tools are often too different to allow for a general test suite. In this paper, we compare fifteen ways of computing derivatives including eleven automatic differentiation tools implementing various methods and written in various languages (C++, F#, MATLAB, Julia and Python), two symbolic differentiation tools, finite differences, and hand-derived computation. We look at three objective functions from computer vision and machine learning. These objectives are for the most part simple, in the sense that no iterative loops are involved, and conditional statements are encapsulated in functions such as {\tt abs} or {\tt logsumexp}. However, it is important for the success of algorithmic differentiation that such `simple' objective functions are handled efficiently, as so many problems in computer vision and machine learning are of this form. Of course, our results depend on programmer skill, and familiarity with the tools. However, we contend that this paper presents an important datapoint: a skilled programmer devoting roughly a week to each tool produced the timings we present. We have made our implementations available as open source to allow the community to replicate and update these benchmarks.

Mario Michael Krell, Matej Kosec, Sergio P. Perez et al.

Effective training of today's large language models (LLMs) depends on large batches and long sequences for throughput and accuracy. To handle variable-length sequences on hardware accelerators, it is common practice to introduce padding tokens, so that all sequences in a batch have the same length. We show in this paper that the variation in sequence lengths in common NLP datasets is such that up to 50% of all tokens can be padding. In less common, but not extreme, cases (e.g. GLUE-cola with sequence length 128), the ratio is up to 89%. Existing methods to address the resulting inefficiency are complicated by the need to avoid cross-contamination in self-attention, by a reduction in accuracy when sequence ordering information is lost, or by customized kernel implementations only valid for specific accelerators. This paper introduces a new formalization of sequence packing in the context of the well-studied bin packing problem, and presents new algorithms based on this formulation which, for example, confer a 2x speedup for phase 2 pre-training in BERT. We show how existing models can be adapted to ensure mathematical equivalence between the original and packed models, meaning that packed models can be trained with existing pre-training and fine-tuning practices.

Benjamin Biggs, Oliver Boyne, James Charles et al.

We introduce an automatic, end-to-end method for recovering the 3D pose and shape of dogs from monocular internet images. The large variation in shape between dog breeds, significant occlusion and low quality of internet images makes this a challenging problem. We learn a richer prior over shapes than previous work, which helps regularize parameter estimation. We demonstrate results on the Stanford Dog dataset, an 'in the wild' dataset of 20,580 dog images for which we have collected 2D joint and silhouette annotations to split for training and evaluation. In order to capture the large shape variety of dogs, we show that the natural variation in the 2D dataset is enough to learn a detailed 3D prior through expectation maximization (EM). As a by-product of training, we generate a new parameterized model (including limb scaling) SMBLD which we release alongside our new annotation dataset StanfordExtra to the research community.

Sebastian Lunz, Yingzhen Li, Andrew Fitzgibbon et al.

Recent work has shown the ability to learn generative models for 3D shapes from only unstructured 2D images. However, training such models requires differentiating through the rasterization step of the rendering process, therefore past work has focused on developing bespoke rendering models which smooth over this non-differentiable process in various ways. Such models are thus unable to take advantage of the photo-realistic, fully featured, industrial renderers built by the gaming and graphics industry. In this paper we introduce the first scalable training technique for 3D generative models from 2D data which utilizes an off-the-shelf non-differentiable renderer. To account for the non-differentiability, we introduce a proxy neural renderer to match the output of the non-differentiable renderer. We further propose discriminator output matching to ensure that the neural renderer learns to smooth over the rasterization appropriately. We evaluate our model on images rendered from our generated 3D shapes, and show that our model can consistently learn to generate better shapes than existing models when trained with exclusively unstructured 2D images.

Benjamin Biggs, Thomas Roddick, Andrew Fitzgibbon et al.

We present a system to recover the 3D shape and motion of a wide variety of quadrupeds from video. The system comprises a machine learning front-end which predicts candidate 2D joint positions, a discrete optimization which finds kinematically plausible joint correspondences, and an energy minimization stage which fits a detailed 3D model to the image. In order to overcome the limited availability of motion capture training data from animals, and the difficulty of generating realistic synthetic training images, the system is designed to work on silhouette data. The joint candidate predictor is trained on synthetically generated silhouette images, and at test time, deep learning methods or standard video segmentation tools are used to extract silhouettes from real data. The system is tested on animal videos from several species, and shows accurate reconstructions of 3D shape and pose.

Amir Shaikhha, Andrew Fitzgibbon, Dimitrios Vytiniotis et al.

We present a system for the automatic differentiation of a higher-order functional array-processing language. The core functional language underlying this system simultaneously supports both source-to-source automatic differentiation and global optimizations such as loop transformations. Thanks to this feature, we demonstrate how for some real-world machine learning and computer vision benchmarks, the system outperforms the state-of-the-art automatic differentiation tools.

Sergey Tulyakov, Andrew Fitzgibbon, Sebastian Nowozin

Deep neural network models trained on large labeled datasets are the state-of-the-art in a large variety of computer vision tasks. In many applications, however, labeled data is expensive to obtain or requires a time consuming manual annotation process. In contrast, unlabeled data is often abundant and available in large quantities. We present a principled framework to capitalize on unlabeled data by training deep generative models on both labeled and unlabeled data. We show that such a combination is beneficial because the unlabeled data acts as a data-driven form of regularization, allowing generative models trained on few labeled samples to reach the performance of fully-supervised generative models trained on much larger datasets. We call our method Hybrid VAE (H-VAE) as it contains both the generative and the discriminative parts. We validate H-VAE on three large-scale datasets of different modalities: two face datasets: (MultiPIE, CelebA) and a hand pose dataset (NYU Hand Pose). Our qualitative visualizations further support improvements achieved by using partial observations.

Qi Liu-Yin, Rui Yu, Lourdes Agapito et al.

We demonstrate the use of shape-from-shading (SfS) to improve both the quality and the robustness of 3D reconstruction of dynamic objects captured by a single camera. Unlike previous approaches that made use of SfS as a post-processing step, we offer a principled integrated approach that solves dynamic object tracking and reconstruction and SfS as a single unified cost function. Moving beyond Lambertian S f S , we propose a general approach that models both specularities and shading while simultaneously tracking and reconstructing general dynamic objects. Solving these problems jointly prevents the kinds of tracking failures which can not be recovered from by pipeline approaches. We show state-of-the-art results both qualitatively and quantitatively.

Cenek Albl, Zuzana Kukelova, Andrew Fitzgibbon et al.

We present new methods for simultaneously estimating camera geometry and time shift from video sequences from multiple unsynchronized cameras. Algorithms for simultaneous computation of a fundamental matrix or a homography with unknown time shift between images are developed. Our methods use minimal correspondence sets (eight for fundamental matrix and four and a half for homography) and therefore are suitable for robust estimation using RANSAC. Furthermore, we present an iterative algorithm that extends the applicability on sequences which are significantly unsynchronized, finding the correct time shift up to several seconds. We evaluated the methods on synthetic and wide range of real world datasets and the results show a broad applicability to the problem of camera synchronization.