Modern deep neural network (DNN) models generally require a huge amount of weight and activation values to achieve good inference outcomes. Those data inevitably demand a massive off-chip memory capacity/bandwidth, and the situation gets even worse if they are represented in high-precision floating-point formats. Effort has been made for representing those data in different 8-bit floating-point formats, nevertheless, a notable accuracy loss is still unavoidable. In this paper we introduce an extremely flexible 8-bit floating-point (FFP8) format whose defining factors - the bit width of exponent/fraction field, the exponent bias, and even the presence of the sign bit - are all configurable. We also present a methodology to properly determine those factors so that the accuracy of model inference can be maximized. The foundation of this methodology is based on a key observation - both the maximum magnitude and the value distribution are quite dissimilar between weights and activations in most DNN models. Experimental results demonstrate that the proposed FFP8 format achieves an extremely low accuracy loss of $0.1\%\sim 0.3\%$ for several representative image classification models even without the need of model retraining. Besides, it is easy to turn a classical floating-point processing unit into an FFP8-compliant one, and the extra hardware cost is minor.

Probabilistic graphical modeling (PGM) provides a framework for formulating an interpretable generative process of data and expressing uncertainty about unknowns, but it lacks flexibility. Deep learning (DL) is an alternative framework for learning from data that has achieved great empirical success in recent years. DL offers great flexibility, but it lacks the interpretability and calibration of PGM. This thesis develops deep probabilistic graphical modeling (DPGM.) DPGM consists in leveraging DL to make PGM more flexible. DPGM brings about new methods for learning from data that exhibit the advantages of both PGM and DL. We use DL within PGM to build flexible models endowed with an interpretable latent structure. One model class we develop extends exponential family PCA using neural networks to improve predictive performance while enforcing the interpretability of the latent factors. Another model class we introduce enables accounting for long-term dependencies when modeling sequential data, which is a challenge when using purely DL or PGM approaches. Finally, DPGM successfully solves several outstanding problems of probabilistic topic models, a widely used family of models in PGM. DPGM also brings about new algorithms for learning with complex data. We develop reweighted expectation maximization, an algorithm that unifies several existing maximum likelihood-based algorithms for learning models parameterized by neural networks. This unifying view is made possible using expectation maximization, a canonical inference algorithm in PGM. We also develop entropy-regularized adversarial learning, a learning paradigm that deviates from the traditional maximum likelihood approach used in PGM. From the DL perspective, entropy-regularized adversarial learning provides a solution to the long-standing mode collapse problem of generative adversarial networks, a widely used DL approach.

In the assembly process of printed circuit boards (PCB), most of the errors are caused by solder joints in Surface Mount Devices (SMD). In the literature, traditional feature extraction based methods require designing hand-crafted features and rely on the tiered RGB illumination to detect solder joint errors, whereas the supervised Convolutional Neural Network (CNN) based approaches require a lot of labelled abnormal samples (defective solder joints) to achieve high accuracy. To solve the optical inspection problem in unrestricted environments with no special lighting and without the existence of error-free reference boards, we propose a new beta-Variational Autoencoders (beta-VAE) architecture for anomaly detection that can work on both IC and non-IC components. We show that the proposed model learns disentangled representation of data, leading to more independent features and improved latent space representations. We compare the activation and gradient-based representations that are used to characterize anomalies; and observe the effect of different beta parameters on accuracy and on untwining the feature representations in beta-VAE. Finally, we show that anomalies on solder joints can be detected with high accuracy via a model trained on directly normal samples without designated hardware or feature engineering.

The latest Deep Learning (DL) models for detection and classification have achieved an unprecedented performance over classical machine learning algorithms. However, DL models are black-box methods hard to debug, interpret, and certify. DL alone cannot provide explanations that can be validated by a non technical audience. In contrast, symbolic AI systems that convert concepts into rules or symbols -- such as knowledge graphs -- are easier to explain. However, they present lower generalisation and scaling capabilities. A very important challenge is to fuse DL representations with expert knowledge. One way to address this challenge, as well as the performance-explainability trade-off is by leveraging the best of both streams without obviating domain expert knowledge. We tackle such problem by considering the symbolic knowledge is expressed in form of a domain expert knowledge graph. We present the eXplainable Neural-symbolic learning (X-NeSyL) methodology, designed to learn both symbolic and deep representations, together with an explainability metric to assess the level of alignment of machine and human expert explanations. The ultimate objective is to fuse DL representations with expert domain knowledge during the learning process to serve as a sound basis for explainability. X-NeSyL methodology involves the concrete use of two notions of explanation at inference and training time respectively: 1) EXPLANet: Expert-aligned eXplainable Part-based cLAssifier NETwork Architecture, a compositional CNN that makes use of symbolic representations, and 2) SHAP-Backprop, an explainable AI-informed training procedure that guides the DL process to align with such symbolic representations in form of knowledge graphs. We showcase X-NeSyL methodology using MonuMAI dataset for monument facade image classification, and demonstrate that our approach improves explainability and performance.

Automatically recommending relevant law articles to a given legal case has attracted much attention as it can greatly release human labor from searching over the large database of laws. However, current researches only support coarse-grained recommendation where all relevant articles are predicted as a whole without explaining which specific fact each article is relevant with. Since one case can be formed of many supporting facts, traversing over them to verify the correctness of recommendation results can be time-consuming. We believe that learning fine-grained correspondence between each single fact and law articles is crucial for an accurate and trustworthy AI system. With this motivation, we perform a pioneering study and create a corpus with manually annotated fact-article correspondences. We treat the learning as a text matching task and propose a multi-level matching network to address it. To help the model better digest the content of law articles, we parse articles in form of premise-conclusion pairs with random forest. Experiments show that the parsed form yielded better performance and the resulting model surpassed other popular text matching baselines. Furthermore, we compare with previous researches and find that establishing the fine-grained fact-article correspondences can improve the recommendation accuracy by a large margin. Our best system reaches an F1 score of 96.3%, making it of great potential for practical use. It can also significantly boost the downstream task of legal decision prediction, increasing the F1 score by up to 12.7%.

Backdoor attack intends to inject hidden backdoor into the deep neural networks (DNNs), such that the prediction of infected models will be maliciously changed if the hidden backdoor is activated by the attacker-defined trigger. Currently, most existing backdoor attacks adopted the setting of static trigger, $i.e.,$ triggers across the training and testing images follow the same appearance and are located in the same area. In this paper, we revisit this attack paradigm by analyzing trigger characteristics. We demonstrate that this attack paradigm is vulnerable when the trigger in testing images is not consistent with the one used for training. As such, those attacks are far less effective in the physical world, where the location and appearance of the trigger in the digitized image may be different from that of the one used for training. Moreover, we also discuss how to alleviate such vulnerability. We hope that this work could inspire more explorations on backdoor properties, to help the design of more advanced backdoor attack and defense methods.

To develop and validate a deep learning–based method for automatic quantitative analysis of lower-extremity alignment. In this retrospective study, bilateral long-leg radiographs (LLRs) from 255 patients that were obtained between January and September of 2018 were included. For training data (n 109), a U-Net convolutional neural network was trained to segment the femur and tibia versus manual segmentation. For validation data (n 40), model parameters were optimized. Following identification of anatomic landmarks, anatomic and mechanical axes were identified and used to quantify alignment through the hip-knee-ankle angle (HKAA) and femoral anatomic-mechanical angle (AMA).

This article is part of the VB Lab Microsoft / NVIDIA GTC insight series. With the rapid pace of change taking place in AI and machine learning technology, it's no surprise Microsoft had its usual strong presence at this year's Nvidia GTC event. Representatives of the company shared their latest machine learning innovations in multiple sessions, covering inferencing at scale, a new capability to train machine learning models across hybrid environments, and the debut of the new PyTorch Profiler that will help data scientists be more efficient when they're analyzing and troubleshooting ML performance issues. In all three cases, Microsoft has paired its own technologies, like Azure, with open source tools and NVIDIA's GPU hardware and technologies to create these powerful new innovations. Much is made of the costs associated with collecting data and training machine learning models.

A part of what we see in science fiction movies will soon become a reality, thanks to artificial intelligence. Every time you saw people talking to holograms in sci-fi movies and thought to yourself "that would be awesome to have", you just might be closer to that future. Smartphones will soon be able to create photorealistic 3D holograms with an AI model developed by a research team at MIT. This system determines the best way to generate holograms from a sequence of input images. This fascinating technology could have applications for VR and AR headsets.

It can take years to learn how to write computer code well. SourceAI, a Paris startup, thinks programming shouldn't be such a big deal. The company is fine-tuning a tool that uses artificial intelligence to write code based on a short text description of what the code should do. Tell the company's tool to "multiply two numbers given by a user," for example, and it will whip up a dozen or so lines in Python to do just that. SourceAI's ambitions are a sign of a broader revolution in software development.