Antoine J.-P. Tixier

École Polytechnique · Computer Science Laboratory

By building on a recently introduced genetic-inspired attribute-based conceptual framework for safety risk analysis, we propose a novel methodology to compute univariate and bivariate construction safety risk at a situational level. Our fully data-driven approach provides construction practitioners and academicians with an easy and automated way of...

We introduce word vectors for the construction domain. Our vectors were obtained by running word2vec on an 11M-word corpus that we created from scratch by leveraging freely-accessible online sources of construction-related text. We first explore the embedding space and show that our vectors capture meaningful construction-specific concepts. We then...

Accepted for publication in Automation in Construction (preprint available online since 16 Aug 2019). This paper significantly improves on, and finishes to validate, the approach proposed in "Application of Machine Learning to Construction Injury Prediction" (Tixier et al. 2016). Like in the original study, we use NLP to extract fundamental attribu...

In light of the increasing availability of digitally recorded safety reports in the construction industry, it is important to develop methods to exploit these data to improve our understanding of safety incidents and ability to learn from them. In this study, we compare several approaches to automatically learn injury precursors from raw constructi...

The present document is a technical report that was delivered by Safety AI as part of a consulting project. It is made publicly available in an anonymized form with expressed consent from the client. The objective of making this report publicly available is twofold: giving future Safety AI's clients a better idea of what they can expect out of a pr...

In this study, we capitalized on a collective dataset repository of 57k accidents from 9 companies belonging to 3 domains and tested whether models trained on multiple datasets (generic models) predicted safety outcomes better than the company-specific models. We experimented with full generic models (trained on all data), per-domain generic models...

Fast and reliable evaluation metrics are key to R&D progress. While traditional natural language generation metrics are fast, they are not very reliable. Conversely, new metrics based on large pretrained language models are much more reliable, but require significant computational resources. In this paper, we propose FrugalScore, an approach to lea...

Inductive transfer learning, enabled by self-supervised learning, have taken the entire Natural Language Processing (NLP) field by storm, with models such as BERT and BART setting new state of the art on countless natural language understanding tasks. While there are some notable exceptions, most of the available models and research have been condu...

Recent work in Dialogue Act (DA) classification approaches the task as a sequence labeling problem, using neural network models coupled with a Conditional Random Field (CRF) as the last layer. CRF models the conditional probability of the target DA label sequence given the input utterance sequence. However, the task involves another important input...

Graph neural networks have recently emerged as a very effective framework for processing graph-structured data. These models have achieved state-of-the-art performance in many tasks. Most graph neural networks can be described in terms of message passing, vertex update, and readout functions. In this paper, we represent documents as word co-occurre...

The number of senses of a given word, or polysemy, is a very subjective notion, which varies widely across annotators and resources. We propose a novel method to estimate polysemy, based on simple geometry in the contextual embedding space. Our approach is fully unsupervised and purely data-driven. We show through rigorous experiments that our rank...

Graph learning is currently dominated by graph kernels, which, while powerful, suffer some significant limitations. Convolutional Neural Networks (CNNs) offer a very appealing alternative, but processing graphs with CNNs is not trivial. To address this challenge, many sophisticated extensions of CNNs have recently been introduced. In this paper, we...

Most graph neural networks can be described in terms of message passing, vertex update, and readout functions. In this paper, we represent documents as word co-occurrence networks and propose an application of the message passing framework to NLP, the Message Passing Attention network for Document understanding (MPAD). We also propose several hiera...

The Hierarchical Attention Network (HAN) has made great strides, but it suffers a major limitation: at level 1, each sentence is encoded in complete isolation. In this work, we propose and compare several modifications of HAN in which the sentence encoder is able to make context-aware attentional decisions (CAHAN). Furthermore, we propose a bidirec...

Accepted for publication in Automation in Construction (Preprint available online since 26 Jul 2019) (1) We propose several methods to automatically learn injury precursors from raw construction accident reports, (2) precursors are learned by multiple supervised models as a by-product of training them at predicting safety outcomes, (3) we experimen...

ive Community Detection is an important Spoken Language Understanding task, whose goal is to group utterances in a conversation according to whether they can be jointly summarized by a common abstractive sentence. This paper provides a novel approach to this task. We first introduce a neural contextual utterance encoder featuring three types of sel...

Graph kernels have been successfully applied to many graph classification problems. Typically, a kernel is first designed, and then an SVM classifier is trained based on the features defined implicitly by this kernel. This two-stage approach decouples data representation from learning, which is suboptimal. On the other hand, Convolutional Neural Ne...

My notes on Deep Learning for NLP.

Graph degeneracy algorithms were recently shown to be very effective at detecting the influential spreaders in a network. However, degeneracy-based decompositions of a graph are unstable to small perturbations of the network structure. At the same time, it is well-known in Machine Learning that the performance of unstable algorithms can be greatly...

We introduce a novel graph-based framework for abstractive meeting speech summarization that is fully unsupervised and does not rely on any annotations. Our work combines the strengths of multiple recent approaches while addressing their weaknesses. Moreover, we leverage recent advances in word embeddings and graph degeneracy applied to NLP to take...

Graph kernels have been successfully applied to many graph classification problems. Typically, a kernel is first designed, and then an SVM classifier is trained based on the features defined implicitly by this kernel. This two-stage approach decouples data representation from learning, which is suboptimal. On the other hand, Convolutional Neural Ne...

We propose and experimentally validate a novel energy-based training module aiming at rapidly improving hazard recognition skills in Civil, Environmental, Architectural and Engineering (CEAE) students. The module is based on the established theory that every construction hazard is fundamentally related to the unwanted release of one or more energy...

The task of graph classification is currently dominated by graph kernels, which, while powerful, scale poorly to large graphs and datasets. Convolutional Neural Networks (CNNs) offer a very appealing alternative. However, feeding graphs to CNNs is not trivial. To address this challenge, many sophisticated extensions of CNNs have recently been propo...

Construction still accounts for a disproportionate number of injuries, inducing consequent socioeconomic impacts. Despite recent attempts to improve construction safety by harnessing emerging technologies and intelligent systems, most frameworks still consider tasks and activities in isolation and use secondary, aggregated, or subjective data that...

The needs to ground construction safety-related decisions under uncertainty on knowledge extracted from objective , empirical data are pressing. Although construction research has considered machine learning (ML) for more than two decades, it had yet to be applied to safety concerns. We applied two state-of-the-art ML models, Random Forest (RF) and...

In the United States like in many other countries throughout the globe, construction workers are more likely to be injured on the job than workers in any other industry. This poor safety performance is responsible for huge human and financial losses and has motivated extensive research. Unfortunately, safety improvement in construction has decelera...

Despite strong advancements in construction safety performance over the past few decades, injuries still occur at an unacceptable rate. Researchers have shown that risk-taking behavior, originating mainly from inaccurate perception and unacceptable tolerance of safety risk, is a significant factor in a majority of construction injuries. Based on ps...

The ability of designers, managers, and workers to identify construction hazards is a fundamental skill that promotes construction safety in practices. Traditionally, construction management programs focus on teaching this topic using the fundamentals of the Occupational Safety and Health Administration and the associated regulations and delivering...