We provide operational consulting in artificial intelligence (AI) for multimedia solutions including feasibility assessment, due diligence, and technology watch. Deraison.AI delivers functional prototypes for clients, while minimizing their exposure to R&D risk.
We mean artificial intelligence in the broadest sense: Mathematics applied with computers also known as statistical learning, pattern recognition, machine learning, signal processing for various media such as sound, image, video, text and of course tabular data.
Our expertise in artificial intelligence was earned through both academic research and corporate experience. Knowing how to combine the best of both worlds is the special signature. We mobilize and create cutting-edge research enhanced by concrete pragmatism to make sure our solutions work in the real world.
Our ambition translates into accomplishments towards artificial intelligence by maintaining close relationships with academia with scientific publications and collaborations, think tanks, companies thanks to research, operational consulting and teaching in business school.
Scientific Publications on Google Scholar
Scientific culture is at the core of our philosophy. We rove across the fields and never hesitate to deploy alternative approaches borrowed from non-AI domains discipline when they work more efficiently.
Welcome to the portfolio of Warith Harchaoui, founder of Deraison.AI, where innovation meets expertise. This section showcases a selection of projects to which Warith has made a significant contribution, both as an ambitious entrepreneur and as a passionate employee.
The focus is not only on the achievements themselves, but also on Deraison.AI's underlying philosophy. The story of each project is told in a way that highlights the results, blending the innovative thinking and strategic mindset behind these successes. From the ambitious ideas to the practical challenges of implementation, these stories provide an insight into Deraison.AI's commitment to creating sustainable business solutions.
“Warith is a Bible in AI”
Annual performance review from Jellysmack, 2022
Jellysmack helps video creators from content production and editing to social media promotion.
At Jellysmack, the team brings best efforts to help the web video creation space with cool projects in computer vision and natural language processing at the service of creators around the world. The mission is twofold :
‣ igniting innovation through applied research prototypes
‣ offering research support to other artificial intelligence/machine learning teams by providing fresh, bibliographic, and pragmatic insights
We aim at inventing and improving the secret sauces, diligently crafting code deliverables, conducting scientific and technology watch, and engaging in internal and external talks.
Proudest achievement? The video representation project, a combination of multiple kernel learning and Nyström approximation. We meticulously gathered and harmonized the finest video representations available from research. With the invaluable insights from our media experts, we did not only infer our own combined similarity but also engineered fixed-sized vector representations for efficient grouping, retrieval, and success prediction estimation.
Additionally, we believe we revolutionize video editing leveraging the work of our hundreds of professional editors, optimizing rush-to-edit and short conversion processes for platforms like YouTube, Facebook, Instagram and TikTok. By analyzing multiple high-level signals and devising various splitting methods, we provide suggestions to editors, ensuring a seamless and creative editing experience minimizing the time spent on tedious non-artistic tasks.
Scientifically, it’s intellectually challenging to deal with videos because this medium is multimodal: image, sound, speech, music and motion. In terms of business, these fields is much rewarding when considering the success of videos and reduced time spent to build which clearly define our pragmatic performance indicators.
“An unlikely but incredible duo!”
Executive MBA course for the Rennes School of Business
Embarking on a transformative journey, Laurent Pantanacce and I carefully craft and deliver a vivid 16-hour Executive MBA course at the Rennes School of Business, titled “Artificial Intelligence for Business”.
This course bridges the chasm between AI academia experts and corporate managers. For managers, CxOs, and tech-savvy AI enthusiasts, the course is designed as a beacon guiding professionals through the labyrinth of AI fundamentals, project management, and real-world application avoiding pitfalls in a sustainable business.
With over 200 slides of constantly updated and tested material, we distill refined knowledge from case studies, practical examples and interactive sessions. The program, which has been rolled out more than three times, is a tapestry weaving together AI basics, challenges, cutting-edge technologies and effective strategies, all aimed at empowering participants to conduct AI projects with confidence and even subtlety for the benefit of customers and sustainable businesses.
With a Ph.D. in applied mathematics and a seasoned tech entrepreneur as instructors, the course is more than a leap of faith; it's a leap to AI mastery and business success.
“Thank you for this great job”
Nathalie Birocheau, founder of Ircam Amplify, 2022
Ircam Amplify's mission is to make the very best in audio research and musical creation from IRCAM labs available to benefit markets with products.
The aim of Ircam Amplify is to blend technology, mathematics and music, creating harmonious solutions that resonate with science and music while striking a chord in the marketplace. In this mission, we looked at maximizing profits on Spotify from our music catalogs.
This involved crafting innovative algorithms to accurately analyze, encode and price to promote old and new tracks. We leverage the dynamics of data, guiding music lovers on a personalized journey of discovery, while taking into account the latest trends and listener success histories.
By intelligently grouping tracks and encoding them adaptively, and by leveraging Ircam's extensive research capital, we enable artificial intelligence systems to appreciate music. We have also successfully experimented with connecting the news feed to improve suggestions.
“Warith's expertise in AI and computer vision has been key to shape the R&D team of VizioSense.”
Maxime Schacht, founder of VizioSense, 2021
VizioSense develops cameras for monitoring and counting that are low-energy, low-bandwidth and respecting privacy by design.
VizioSense diligently innovates to launch low-energy, low-bitrate cameras that excel at counting cars on roads and parking lots, but also people in stores for marketing analysis purposes.
Our expertise in AI and computer vision served as the foundation for the team's prosperity and evolution. Since then, VizioSense has not only acquired customers but also earned solid trust - a testament to the relentless pursuit of excellence celebrated from Lille to Paris.
VizioSense has raised millions of euros, fuelling its expansion into the smart camera sector with ingeniously energy- and bandwidth-efficient products. It's a real source of pride to witness their success long after our small contribution.
“A fascinating (and passionate) work which shows a beautiful hindsight on the field.”
Erwann Le Pennec, Professor at École Polytechnique and Ph.D. jury referee, 2021
MAP5 is the Applied Math lab of Université de Paris that hosts Warith Harchaoui studying Generative and Discriminative AI for 4 wonderful years.
During four life-changing years at MAP5 (2016-2020) within Université de Paris's applied math lab, Warith delved into the enthralling realm of Generative and Discriminative AI. Engaging deeply with neural networks and deep learning, Warith repurposed these potent tools. His work included pioneering clustering techniques with GANs, adept at organizing vast datasets, like categorizing myriad pixels in extensive photo collections. Warith also explored unsupervised feature importance, revealing distinctive dataset attributes without labels.
Additionally, the academic and corporate work was about methods to assess automated decision reliability, providing a crucial statistical approach for applications needing human oversight. This exploration merges some mathematics from optimal transport and neural networks, which helps reevaluating data representation. Thanks to the warmth of MAP5 and the pugnacity of Oscaro, Warith's journey modestly contributed to the field which is immense for him.
Afterwards, bringing home publications from top research conferences and journals brings a mixed feeling of pride, vanity and recognition, but also of course the emboldened drive to fight the next research battle in companies.
“This cooperation with the university brought great value to the company I founded, Oscaro, with the product Cerbero.”
Pierre-Noël Luiggi, founder of Oscaro.com, 2020
Oscaro.com is the worldwide leading e-commerce for car parts.
At Oscaro.com, we participated in innovative initiatives as R&D Data Scientists, seamlessly blending deep learning and artificial intelligence to craft robust solutions. Our work was pivotal in implementing cutting-edge credit card fraud detection mechanisms, significantly mitigating risks and safeguarding transactions. Concurrently, we delved into demand forecasting, utilizing deep learning to accurately predict and adeptly respond to market dynamics, thereby driving efficiency and customer satisfaction.
Credit card fraud detection was a good example of asking ourselves the right questions: at the time, companies used to count errors, our approach consisted in counting the company's money loss. All systems in AI and beyond make mistakes: our job is to handle them in the most meaningful way. Thinking this way completely modifies the way you measure algorithm performances and also the way algorithms are rewarded and thus trained. It's also beneficial for management as the whole chain of command, even without AI knowledge, now understands what is optimized: the company's money. It turns out this approach made our career much easier with companies and clients.
Our role also encompassed AI technology watch, where we diligently monitored and analyzed emerging trends, research, and companies in the AI sphere, ensuring Oscaro.com stayed ahead in the competitive landscape. This commitment to excellence was further underscored by our pursuit of a Ph.D., generously funded from September 2016 to October 2020. Balancing from academic to work environments enriched our AI points of view, both feeding the common intellectual thirst.
Artificial Intelligence should be understood as an upheaval comparable to agriculture and sedentary life, the invention of writing or even printing. Those turning points altered our very relationship to the world.
All activities are now being affected by AI. For our everyday life, artificial intelligence is the automated emergence of certain aspects of natural intelligence such as learning, decision making, adaptation, prediction etc. In practice, artificial intelligence is the science that allows machines to execute tasks from examples instead of instructions.
deraison.ai is committed to contributing to the international research scientific endeavor through blogs and peer-reviewed publications.
WTOT - Weighted Transformation Optimal Transport
We have created methods for geometrically connecting two piles of points, which is useful when these points are supposed to match. Our application example concerns the study of messenger RNA in genetics to understand Huntington's disease.
Thi Thanh Yen Nguyen, Warith Harchaoui, Lucile Mégret, Cloe Mendoza, Olivier Bouaziz, Christian Neri and Antoine Chambaz
Clustering is the process of making groups among objects: similar within a group, dissimilar between two groups. The human mind naturally likes to form groups when it comes to customers, genes, fruits, animals and so on. The GEMINI project exploits artificial neural networks in a novel way for this large-scale task.
Louis Ohl, Pierre-Alexandre Mattei, Charles Bouveyron, Warith Harchaoui, Mickaël Leclercq, Arnaud Droit and Frederic Precioso
Learning Representations using Neural Networks and Optimal Transport
From 2016 to 2020, this Ph.D. work focused on three key areas:
‣ Clustering: grouping large-scale data in number and dimension ;
‣ Importance of data features: identifying representative and discriminating features in data without the need for manual labels ;
‣ Confidence level estimation in automated decisions: creating statistical methods to assess the reliability of automated decisions, which is crucial in industry, security and healthcare.
The originality of this research lies in the way it combines the mathematical fields of optimal transportation and neural networks in particular with generative artificial intelligence, which was in its infancy.
Warith Harchaoui holds a Ph.D. in applied mathematics and is deeply engaged in the intersection of artificial intelligence (AI) in academia and in companies. He commenced his academic journey at École Normale Supérieure de Cachan, where he received his M.Sc. in 2008. His initial forays into the industry revolved around computer vision applications in startups and global companies.
He extended his academic pursuits at École Normale Supérieure de Paris, contributing to research at the globally recognized Willow laboratory. His scholarly interests encompass a range of topics within image, sound, video, and text processing, a domain where he has consistently leveraged mathematical models for innovative solutions. From 2014 to 2020, Warith seamlessly transitioned into a corporate role in data science, working with the e-commerce leader in the automotive parts sector Oscaro.com. Balancing industry demands with academic rigor, he completed his Ph.D. research between 2016 and 2020 at the MAP5 Lab of Université de Paris.
Ph.D. defense on Learning Representations using Neural Networks and Optimal Transport
From 2021 to 2024, Warith worked at Jellysmack, among the most innovative companies in 2022 et 2023 (according to FastCompany and Digiday), where he helped web video creators every step of the way, from content production and editing to social media promotion. With an infectious glee, he honed his ability to spark innovation with prototypes and act as a scientific reference, thanks to a voracious appetite for reading publications. Warith's career has benefited from his pleasure in sharing and enriching his scientific knowledge through internal and external seminars wherever he has worked. Taking a genuine interest in everyone's work is his real way to work with everyone!
Today, Warith continues to contribute to leading scientific conferences and journals of AI (including Neurips 2022 and Journal of the Royal Statistical Society 2023). His confirmed know-how encouraged him to embrace new dreams: bringing his expertise from artificial intelligence into the real world at the service of companies with his friend, collaborator and mentor in business Laurent Pantanacce.