Offre en lien avec l’Action/le Réseau : DOING/– — –

Laboratoire/Entreprise : SAMOVAR – Télécom SudParis
Durée : 6 mois
Contact : romerojulien34@gmail.com
Date limite de publication : 2023-09-02

Contexte :
Given a text, it is possible to extract from it knowledge in the form of subject-predicate-object triples, where all components of the triples can be found in the text. This is called Open Information Extraction (OpenIE). For example, from the sentence “The fish swims happily in the ocean”, we can extract the triple (fish, swims, in the ocean). By gathering many of these statements, we obtain an Open Knowledge Base (OpenKB), with no constraints on the subjects, the predicates, and the objects.

Then, this OpenKB could be used for question answering (QA). There have been many approaches that target QA over non-open KBs. These approaches vary from crafting query templates that, once filled in, will be used to query the KB, to neural models, where the goal is to represent the question and the possible answers as latent vectors, where the correct answer should be close in the embedding space to the question~cite{bordes2014question}. In this project, we will focus on neural models, particularly knowledge graph embeddings, i.e., continuous representations for the entities and relations that can generally capture relevant information about the graph’s structure.

The current way KB embeddings are computed raises two main challenges:
* Each entity and relation must be seen enough times during training so the system can learn relevant embeddings. The training is done taking edges information into account, so the entity or relation must be part of a sufficiently large number of edges.
* The textual representation of the verbal and noun phrases of the relations, subjects, and objects should be considered.

For example, a recent approach, MHGRN, computes embeddings by using a modified graph neural network architecture. This architecture, however, does not take into account the textual representation of relations.
A better approach is CARE, that relies on two main ideas. First, it clusters the subjects and objects and creates an unlabelled edge between entities in the same cluster. That partially reduces the problem of the entities connected to a small number of edges, by leveraging the connection with better connected entities. Then, it computes embeddings for the relations using GLOVE (word embeddings) and GRUs (recurrent neural networks). We believe that the approach in CARE could be improved by considering more modern neural architectures using message-passing algorithms and integrating the textual representation of predicates, objects, and subjects. In addition, we will investigate if the clustering step is necessary, as it can bring a bias for one important downstream application of KB embeddings: canonicalization, the task of finding a representative for a set of nodes or edges.

In this project, we will improve open KB embedding methods by:
* Exploring state-of-the-art neural architectures and language models.
* Integrating textual representations of the subject, predicate, and object.
* Investigating if clustering before embedding computation is necessary.
* Integrating embeddings into question-answering models.

Sujet :
Given a text, it is possible to extract from it knowledge in the form of subject-predicate-object triples, where all components of the triples can be found in the text. This is called Open Information Extraction (OpenIE). For example, from the sentence “The fish swims happily in the ocean”, we can extract the triple (fish, swims, in the ocean). By gathering many of these statements, we obtain an Open Knowledge Base (OpenKB), with no constraints on the subjects, the predicates, and the objects.

Then, this OpenKB could be used for question answering (QA). There have been many approaches that target QA over non-open KBs. These approaches vary from crafting query templates that, once filled in, will be used to query the KB, to neural models, where the goal is to represent the question and the possible answers as latent vectors, where the correct answer should be close in the embedding space to the question~cite{bordes2014question}. In this project, we will focus on neural models, particularly knowledge graph embeddings, i.e., continuous representations for the entities and relations that can generally capture relevant information about the graph’s structure.

The current way KB embeddings are computed raises two main challenges:
* Each entity and relation must be seen enough times during training so the system can learn relevant embeddings. The training is done taking edges information into account, so the entity or relation must be part of a sufficiently large number of edges.
* The textual representation of the verbal and noun phrases of the relations, subjects, and objects should be considered.

For example, a recent approach, MHGRN, computes embeddings by using a modified graph neural network architecture. This architecture, however, does not take into account the textual representation of relations.
A better approach is CARE, that relies on two main ideas. First, it clusters the subjects and objects and creates an unlabelled edge between entities in the same cluster. That partially reduces the problem of the entities connected to a small number of edges, by leveraging the connection with better connected entities. Then, it computes embeddings for the relations using GLOVE (word embeddings) and GRUs (recurrent neural networks). We believe that the approach in CARE could be improved by considering more modern neural architectures using message-passing algorithms and integrating the textual representation of predicates, objects, and subjects. In addition, we will investigate if the clustering step is necessary, as it can bring a bias for one important downstream application of KB embeddings: canonicalization, the task of finding a representative for a set of nodes or edges.

In this project, we will improve open KB embedding methods by:
* Exploring state-of-the-art neural architectures and language models.
* Integrating textual representations of the subject, predicate, and object.
* Investigating if clustering before embedding computation is necessary.
* Integrating embeddings into question-answering models.

Profil du candidat :
The intern should be involved in a master’s program and have a good knowledge of machine learning, deep learning, natural language processing, and graphs. A good understanding of Python and the standard libraries used in data science (scikit-learn, PyTorch, pandas, transformers) is also expected. In addition, a previous experience with graph neural networks would be appreciated.

Formation et compétences requises :
The intern should be involved in a master’s program and have a good knowledge of machine learning, deep learning, natural language processing, and graphs. A good understanding of Python and the standard libraries used in data science (scikit-learn, PyTorch, pandas, transformers) is also expected. In addition, a previous experience with graph neural networks would be appreciated.

Adresse d’emploi :
Palaiseau

Document attaché : 202302091340_internship_openie-1.pdf

Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : LAMSADE – PSL Research University – Universit{
Durée : 4 à 6 mois
Contact : maude.manouvrier@lamsade.dauphine.fr
Date limite de publication : 2023-02-09

Contexte :
{Data dependencies : relationships or connections between different variables in a dataset. Understanding these dependencies is crucial and has a number of applications.

{Data profiling for Machine Learning: Understanding data dependencies is critical for creating accurate and effective machine learning models. The quality of the input data has a direct impact on the accuracy of the model, and understanding data dependencies helps ensure that the data is suitable for use in machine learning.

Data mining: Data dependencies can help you identify patterns and relationships in the data that may not be immediately obvious. These patterns can be used to make predictions and classify data, making it useful in various data mining tasks such as association rule mining and clustering.

Sujet :
This internship will build upon the recent research in data dependency mining in dynamic settings. As a member of a dynamic team, the student will be exploring innovative ways to compute data dependencies in situations where the data is transformed through a data preparation pipeline. The goal is to assess the impact of this preparation process on the dependencies within the data, as well as its overall quality.

The subject of data dependencies is a critical and fascinating aspect of machine learning and AI, providing students with the opportunity to gain practical skills and explore cutting-edge technologies that are shaping the future of the field. The demand for professionals with skills in machine learning and AI is growing rapidly, and understanding data dependencies is a valuable skill for anyone looking to build a career in this field in both academia and industry. On this point, it is worth noting that the internship is likely to lead to a PhD on a related topic.

Profil du candidat :
We seek for excellent and highly motivated student with a background in Computer Science
having good knowledge of database theory and good programming skills (Python or Java).

Please send the following material in a single PDF document before February 20th, 2023:
– fully detailed CV,
– academic records (master’s degree or equivalent),
– recommendation(s) and supporting letter(s).

Formation et compétences requises :
Background in Computer Science
Good knowledge of database theory and good programming skills (Python or Java).

Adresse d’emploi :
LAMSADE – PSL Research University – Universit{‘e} Paris-Dauphine, Paris, France

Document attaché : 202302091126_IntershipLamsadeDataDependencieInPiplines.pdf

Offre en lien avec l’Action/le Réseau : – — –/Innovation

Laboratoire/Entreprise : The candidate will be either located at CRAN, Nanc
Durée : up to 6 months
Contact : zniyed@univ-tln.fr
Date limite de publication : 2023-05-30

Contexte :
Many imaging applications rely on the acquisition, processing and analysis of 3D or 4D vectorial data pixels: this includes notably color imaging (red, blue and green channels) or polarimetric imaging (4D Stokes parameters at each pixel). Such multichannel data is often represented using quaternions – a generalization of complex numbers in four dimensions – in order to simplify expressions and leverage unique geometric and physical insights offered by this algebraic representation. Therefore, datasets of color or polarimetric images can be viewed as a collection of quaternion-valued matrices, which form multidimensional quaternion arrays – also called quaternion tensors.

Sujet :
The aim of this internship is to demonstrate the potential of quaternion tensor decompositions for learning features from databases of color and polarimetric images. Quaternion tensor decompositions have only been introduced recently [1]. They generalize usual tensor decompositions
[2] to the quaternion field. The candidate will take advantage of the algorithms proposed in [1]. He / she will focus on two main cases of uses of quaternion tensor decompositions (Canonical Polyadic and Tucker) to

1. learn features from a standard color image database (such as ImageNET)
2. perform source separation on polarimetric hyperspectral data

One key complementary objective will be to benchmark performances of quaternion tensor decompositions
against standard real-domain tensor decompositions.

Profil du candidat :
The candidate should have good writing and oral communication skills.

Formation et compétences requises :
He/she should be enrolled in a M1/M2R or engineer diploma in one or more of the following fields: signal and image processing, machine learning, applied mathematics.

Adresse d’emploi :
Depending on his/her preferences, the candidate will be either located at CRAN, Nancy or either at LIS, Seatech, Toulon.

Document attaché : 202302081818_projet.pdf

Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : LIP6 (Sorbonne Université / CNRS)
Durée : 6 months
Contact : lionel.tabourier@lip6.fr
Date limite de publication : 2023-05-30

Contexte :
PPI (protein-protein interaction) networks represent interactions between proteins within a living organism. PPI network maps are incomplete because checking the existence of each relationship demands specific experiments, it is therefore desirable to have means to select the most probable interactions. Recent works brought to light the fact that link prediction approaches are relevant to detect interactions between proteins.

Sujet :
The approaches in question are unsupervised, however there exist supervised methods which have been designed for analogous problems in other contexts. We think that it is possible to adapt such methods to the context of PPI networks. By defining adequate graph features – particularly specific graph motifs – in order to achieve the learning, it would be possible to improve significantly the predictive power of these methods. The purpose of the internship is to design and apply such prediction methods.

The developed methods will be trained and validated using several networks comprised of 5 000 – 18 000 proteins (nodes) establishing between 20 000 and more than 2 million experimentally validated interactions (edges) coming from reference PPI resources, namely the STRING database, the BioGRID, and the Human Reference Interactome.

Profil du candidat :
This internship is preferably directed at Master 2 students with a background in computer science or bioinformatics.

Formation et compétences requises :
Good coding skills are requested for the internship, knowledge of a widely-used language in learning, such as python, is preferable but not mandatory. An open-mind to interdisciplinary applications is certainly a plus.

Adresse d’emploi :
LIP6, 4 Place Jussieu, 75005 Paris

Document attaché : 202302081543_Stage_Link_Pred.pdf