Conditional random fields (CRF) are widely used for sequence labeling such as Named Entity Recognition (NER) problems. Most CRFs, in Natural Language Processing (NLP) tasks, model the dependencies between predicted labels without any consideration for the syntactic specificity of the document. Unfortunately, these approaches are not flexible enough to consider grammatically rich documents like patents. Additionally, the position and the grammatical class of the words may influence the text’s understanding. Therefore, in this paper, we introduce SynCRF which considers grammatical information to compute pairwise potentials. Syn-CRF is applied to TRIZ (Theory of Inventive Problem Solving), which offers a comprehensive set of tools to analyze and solve problems. TRIZ aims to provide users with inventive solutions given technical contradiction parameters. SynCRF is applied to mine these parameters from patent documents. Experiments on a labeled real-world dataset of patents show that SynC RF outperforms state-of-the-art and baseline approaches.

In recent years, Deep Learning methods have become very popular in Natural Language Processing (NLP), especially transformer-based architecture. NLP domain requires a high volume of annotated data to work. Unfortunately, obtaining high-quality and voluminous labeled data is expensive and time-consuming. One promising method which has singled out for its performance in the context of data deficiency is semi-supervised learning with Generative Adversarial Networks (GAN). In this paper, we propose a new approach called PaGAN which is a combination of a document classifier and a sentence-level classifier inside a GAN for patent documents understanding. The idea is to mine the patent’s motivating problem (aka contradiction in TRIZ domain) which is fundamentally important to understand the underlying invention and its originality. PaGAN is applied and evaluated on a real-world dataset. Experiments show outperforming results of PaGAN comparatively to baseline approaches.

TRIZ theory is considered as a kind of innovative theory, which mainly functions in solving contradiction and aims to improve engineers creativity. It may reason on different support such as web and scientific documents. Most of the current knowledge, especially in industrial domains lies in patents. They are, nevertheless, an underused resource because of their complexity, their length and the need of domain knowledge to make use of patent. In this paper, we introduce an innovative application of deep learning to mine knowledge from patents. We show that our summarization based approach, called SummaTRIZ1, enables the extraction of these contradictions essential for TRIZ problem solving engine. A BERT based summarization approach is introduced to retain contradiction sentences. Our approach is experimentally evaluated on a real data set showing the performance of SummaTRIZ.

We present a temporal Bayesian filter for semantic segmentation of a video sequence. Each pixel is a random variable following a discrete probabilistic distribution function representing possible semantic classes. Bayesian filtering consists in two main steps: 1) a prediction model and 2) an observation model (likelihood). We propose to use a datadriven prediction function derived from a dense optical flow between images t and t + 1 achieved by a deep neural network [1]. Moreover, the observation function uses a semantic segmentation network. The resulting approach is evaluated on the public dataset Cityscapes. We show that using the temporal filtering increases the accuracy of the semantic segmentation.