St Germain en Laye, 20 janvier 2025.

Nous contacter pour inscription spontanée : https://nexyad.net/Automotive-Transportation/contact-nexyad/

Les dix points suivants seront abordés :

1. Introduction à l’IA et à ses concepts fondamentaux

• Définir l’IA et ses sous-domaines : apprentissage supervisé, apprentissage non supervisé, réseaux neuronaux, machine learning, deep learning, etc.
• Comprendre la différence entre l’IA, l’automatisation et l’analytique avancée.
• Prendre conscience des tendances actuelles en IA et de son évolution.

2. Impact de l’IA sur les entreprises et les secteurs

• Identifier les domaines d’application de l’IA dans différents secteurs (industrie, santé, finance, marketing, etc.).
• Exemples de transformation d’entreprises grâce à l’IA : amélioration de l’efficacité, personnalisation des services, innovation produit, prise de décision prédictive.
• Étudie de cas d’usage concrets et des réussites.

3. La gouvernance de l’IA et les enjeux éthiques

• Enjeux éthiques, la transparence, la protection des données et biais dans les algorithmes.
• Nécessité de réguler l’utilisation de l’IA dans un cadre légal et éthique.
• Analyser les implications sur la confidentialité et la sécurité des données.

4. Stratégie de transformation numérique et d’intégration de l’IA

• Comment définir une stratégie d’adoption de l’IA au sein de l’organisation.
• Identifier les processus qui peuvent être améliorés ou automatisés par l’IA.
• L’importance d’une culture de données pour réussir une transformation numérique.
• Gérer le changement au sein de l’entreprise en intégrant l’IA dans les équipes et les opérations.

5. Prendre des décisions stratégiques grâce à l’IA

• L’utilisation de l’IA pour l’analyse prédictive et l’aide à la prise de décision.
• L’intégration de l’IA dans les outils de gestion et les systèmes d’information.
• Comment l’IA peut améliorer la prise de décision rapide et éclairée.

6. Compétences et talents nécessaires pour implémenter l’IA

• Identifier les compétences techniques nécessaires au sein des équipes (data scientists, ingénieurs en IA, etc.).
• Importance de la formation continue et du recrutement de talents spécialisés en IA.
• Collaboration entre équipes techniques et non techniques pour réussir l’adoption de l’IA.

7. Risques et défis liés à l’IA

• Comprendre les risques technologiques, En particulier, enjeux de confidentialité, de licences et de droit d’utilisation., organisationnels et financiers associés à l’adoption de l’IA.
• Gestion de l’acceptation de l’IA par les collaborateurs et les parties prenantes.
• Les défis liés à l’intégration des systèmes IA avec les systèmes existants.

8. Retour sur investissement (ROI) de l’IA

• Mesurer les bénéfices réels de l’adoption de l’IA pour l’entreprise : réduction des coûts, amélioration de la productivité, satisfaction client accrue, etc.
• Les clés pour maximiser le ROI tout en minimisant les coûts et risques associés à l’implémentation.

9. Les outils et technologies d’IA à connaître

• Survol des principales plateformes et outils d’IA disponibles : outils d’analyse de données, plateformes cloud, logiciels de machine learning.
• Introduction aux outils sans code/no-code pour l’IA, utiles dans les entreprises non-techniques.

10. Planification de la roadmap IA et pilotage du projet

• Comment établir une feuille de route pour le déploiement de l’IA dans l’entreprise.
• Suivi et gestion des projets IA : gestion de la performance, mesure de l’efficacité, gestion des ressources.
• La nécessité d’une approche agile dans l’implémentation de l’IA.

Nous contacter pour inscription spontanée : https://nexyad.net/Automotive-Transportation/contact-nexyad/

St Germain en Laye, January 29th 2025.

About: an overview on generative AI tools for working with text data, including creating and analyzing data from a variety of sources; discussion of benefits, capabilities, and challenges including how to assess bias. Hands-on exercises focused on real-world examples using text data.

Instructor Dr. Mark Hansen was the inaugural director for the David and Helen Gurley Brown Institute of Media Innovation and is a data scientist at Columbia University working at the intersection of data, art and technology. His work has appeared in the Museum of Modern Art in New York, the Whitney Museum, and the lobby of the New York Times.

St Germain en Laye, January 28th 2025.

Exciting new research explores the role of Artificial Intelligence (AI) in nutrition! A recent scoping review highlights AI’s growing potential in dietary assessment, lifestyle interventions, and disease prediction. However, ethical considerations and data biases remain key challenges. The study emphasizes the need for further research to unlock AI’s full potential in revolutionizing human nutrition.

See the resaerch paper here: The Role of Artificial Intelligence in Nutrition Research: A Scoping Review – PMC

#AI #nutrition #healthcare #research #Nexyad

St Germain en Laye, January 27th 2025.

DeepSeek was developed for less than $6 million, raising concerns about American dominance in the AI field. The chatbot uses an open-source model, which has been criticized by some experts. American restrictions on the sale of advanced chip technologies to China have pushed Chinese developers to innovate, leading to AI models requiring less computing power.

DeepSeek was founded in 2023 by Liang Wenfeng, who collected banned Nvidia chips to create his model. Liang recently expressed his surprise regarding the sensitivity of the prices related to his product.

See Nexyad AI page: Artificial Intelligence: We bring Solutions to your Problems

#AIRevolution #DeepSeek #ChabotChallenge #TechStocks #AIDominance #InnovationInAI #ChineseTech #Nexyad

St Germain en Laye, January 23nd 2025.

Stargate will be the company created by tech giants OpenAI, SoftBank, and Oracle to enhance artificial intelligence (AI) infrastructure in the U.S.

The announcement was made at the White House, where President Donald Trump called this initiative the “largest AI infrastructure project in history.” The companies plan to invest an initial $100 billion, with long-term projections reaching up to $500 billion, aiming to generate around 100,000 new jobs.

Stargate’s objective is to establish both physical and virtual infrastructure essential for AI, including data centers across the nation. OpenAI’s CEO Sam Altman emphasized the crucial nature of robust infrastructure for AI, especially in the context of an intensifying global competition with China for AI supremacy. The initiative aims not only to bolster U.S. technological capabilities but also to secure economic benefits and create job opportunities domestically.

Despite the ambitious goals, historical context reveals skepticism about such announcements. The article recalls Trump’s past declaration regarding a $10 billion Foxconn factory in Wisconsin, which ultimately did not materialize as promised. This suggests that while the Stargate project has significant potential, its success will depend on effective execution and follow-through.

The Stargate project could represent a critical step in the U.S. solidifying its position in the competitive tech landscape, especially concerning AI. However, the credibility of such large-scale promises hinges on the commitment of all stakeholders to see through the plans. Tracking the project’s progress will be essential, as past experiences with similar initiatives raise questions about their reliability and long-term impact. Emphasizing collaboration between the private sector and government, while addressing transparency and regulation, could enhance the chances of success for Stargate and reshape the future of AI in America.

Read CNN article: Stargate: Trump announces a $500 billion AI infrastructure investment in the US | CNN Business

#AIInfrastructure #StargateProject #ArtificialIntelligence #TechInvestment #OpenAI #SoftBank #Oracle #JobCreation #AIEconomy #USInnovation #Nexyad

St Germain en Laye, January 21st 2025.

According to recent developments highlighted in the research on GenCast, Google DeepMind’s AI weather forecasting model has shown to provide better forecasts than some leading operational models, including the European Centre for Medium-Range Weather Forecasts’ ensemble model (ENS).

GenCast can generate accurate forecasts up to 15 days in advance and does so in a fraction of the time compared to traditional models, which often take hours to provide similar forecasts. The model outperformed ENS in 97% of the measures used to evaluate its probabilistic forecasts, particularly excelling at predicting extreme weather events such as hurricanes and heatwaves.

This indicates that GenCast has the potential to deliver more reliable and faster weather forecasts than established agencies, marking a significant advancement in the field of meteorology driven by machine learning and AI technology.

Read Nature publication here: DeepMind AI weather forecaster beats world-class system

#AIWeatherForecasting #DeepMind #GenCast #MachineLearning #WeatherPrediction #ClimateTech #ExtremeWeather #Meteorology #ArtificialIntelligence #OpenScience #AI #Nexyad

St Germain en Laye, January 17th 2025.

Here an interesting article about AI in political campaign by Christina LaChapelle and Catherine Tucker for Brennan Center:

The rise of generative AI presents both opportunities and challenges for political campaigns. AI offers cost-effective, highly targeted advertising, leveling the playing field for smaller campaigns. However, this technology’s potential for producing inaccurate, biased, or unoriginal content poses significant risks, especially in the absence of regulation. Campaigns must prioritize human oversight and fact-checking to ensure responsible use and maintain voter trust. While market forces may eventually favor higher-quality AI-generated ads, proactive measures are necessary to prevent the misuse of this powerful technology.

Headlines:

How Can AI Make Political Advertising More Powerful?

• Targeting Specific Audiences
• Empowering Less-Resourced Campaigns
• Improving Ad Effectiveness

What Are the Risks of AI Use in Political Advertising?

• Falsehoods and Empty Promises
• Biases
• Ignorance of Certain Topics
• Generic Language

What to Do: Can Market Forces Help?

Discover full article on Brennan Center website and many links: Generative AI in Political Advertising | Brennan Center for Justice

#AIinPolitics #PoliticalAdvertising #GenerativeAI #CampaignTech #Election2024 #Misinformation #Deepfakes #DigitalCampaigning #PoliticalTech #AIethics #AI #Nexyad

St Germain en Laye,  January 16th 2025.

This study investigated the application of artificial intelligence (AI) to improve the diagnosis of anorectal motility disorders, a significant healthcare challenge due to the complexity and limited accessibility of high-resolution anorectal manometry (HR-ARM). The researchers developed and validated a machine learning (ML) model to automatically detect and differentiate various motility patterns based on HR-ARM data.

The study utilized a large dataset (701 HR-ARM exams) from a tertiary care center, classified according to the standardized London Classification. The data was split into training (80%) and testing (20%) sets for model development and evaluation. Multiple ML algorithms were tested, and the Light Gradient Boosting Machine (LGBM) classifier demonstrated superior performance, achieving an accuracy of 87% in identifying disorders of anal tone and contractility. Furthermore, individual ML models exceeded 90% accuracy in differentiating specific disorder subtypes (e.g., anal hypotension with normal contractility, anal hypertension).

The findings underscore the potential of AI to address key limitations of HR-ARM, including its complex data analysis, limited accessibility, and inter-observer variability in interpretation. By automating the detection of motility patterns, the AI model offers a promising solution for improving diagnostic accuracy, efficiency, and accessibility of HR-ARM, leading to more timely and effective management of anorectal functional disorders.

The study acknowledges limitations, primarily the single-center nature of the dataset, and suggests future research to incorporate data from multiple centers and diverse patient populations to enhance the generalizability and robustness of the AI model. The development of explainable AI models is also highlighted as a crucial next step to increase transparency and build trust in AI-driven diagnostics. This research represents a significant advancement in the application of AI to gastroenterology and offers a pathway to improve patient care in the management of anorectal disorders. The successful application of AI in this context lays the groundwork for broader adoption of AI-assisted diagnostics in other areas of gastroenterology and beyond.

Read Nature paper.

#AIinHealthcare #ArtificialIntelligence #MachineLearning #AnorectalManometry #Gastroenterology #MotilityDisorders #HRARM #MedicalTechnology #Nexyad #DataAnalysis#PrecisionMedicine

St Germain en Laye, January 15th 2025.

This research investigated the use of ensemble machine learning techniques to predict obesity risk using lifestyle data. The study employed a range of algorithms from three ensemble learning categories: boosting (XGBoost, Gradient Boosting, CatBoost), bagging (Bagged Decision Tree, Random Forest, Extra Trees), and voting (Logistic Regression, Decision Tree, Support Vector Machine). A publicly available dataset containing lifestyle information and obesity levels was preprocessed to handle missing values and outliers before model training. Hyperparameter tuning and feature selection (using recursive feature elimination) were performed to optimize model performance.

The results showed that XGBoost achieved the highest accuracy (98.1%), precision (97.5%), recall (97.5%), and F1-score (96.5%), along with a perfect AUC-ROC score of 100%. Weight, height, and age were consistently ranked as the most significant predictors of obesity risk across various models. Other factors like family history, diet, physical activity, and technology use also contributed to the prediction, although with varying degrees of influence depending on the model. The study also analyzed the performance of the other ensemble methods, revealing that boosting techniques generally outperformed bagging and voting in this specific task. The confusion matrices, precision, recall, F1-score, and AUC-ROC curves for each model provided a detailed analysis of performance across different obesity levels.

The authors compared their results with those of previous studies, highlighting the superior performance of their XGBoost model. However, they also acknowledged limitations, such as the use of a synthesized dataset which might limit the generalizability of the findings. Future research directions suggested by the authors include using larger, more diverse datasets, incorporating additional relevant features, exploring deep learning methods, and enhancing model interpretability through techniques like SHAP values. Overall, the study demonstrates the potential of ensemble learning and specifically XGBoost for accurate and efficient prediction of obesity risk, paving the way for improved early detection and intervention strategies.

Read the paper:

#ObesityPrediction #MachineLearning #EnsembleLearning #AIinHealthcare #HealthTech #PredictiveModeling #ObesityRisk #LifestyleData #DataScience #XGBoost #AI #Nexyad

St Germain en Laye, January 14th 2025.

Nexyad offers fleet management solutions using AI-powered onboard real-time driver assistance. Our system, SafetyNex, collects valuable driving data 20 times per second, assessing driver prudence levels and identifying risky behaviors. This data helps pinpoint instances of risky driving, including location, time, cause of risk, and driving style (eco, smooth, harsh, or normal).

The system provides real-time feedback to drivers, offering alerts and anticipatory warnings before dangerous situations, ultimately improving driver safety and reducing accidents. Post-trip, drivers receive personalized safety scores and coaching for continuous improvement. For fleet managers, this translates to reduced accident rates, lower operational costs (including TCO, driver injuries, and downtime), and an improved company image through enhanced social responsibility.

Nexyad offers two solutions: a standalone smartphone app (Motiv AI) and an SDK for integration into existing devices (telematics, dashcams, etc.). Both leverage SafetyNex’s core technology for analyzing driving behavior in relation to road context (curves, intersections, pedestrian crossings, etc.), using artificial intelligence to identify and address potential hazards. The data recorded includes a wide range of variables, enabling comprehensive analysis and reporting. The Motiv AI app provides drivers with feedback, while the SDK facilitates seamless integration with existing fleet systems.

Contact Nexyad

#Nexyad #FleetSafety #DriverMonitoring #AI #ArtificialIntelligence #RoadSafety #DriverAssistance #Telematics #ADAS #SmartDriving #FleetManagement #SafetyTechnology #SafetyNex #BYOD

St Germain en Laye, January 13th 2025.

While AI augments ATCO capabilities through improved data analysis and conflict prediction, final decision-making remains with the human controller. IFATCA advocates for a human-centered approach, designing intuitive interfaces and ensuring robust safety mechanisms to prevent over-reliance on automated systems. Continuous training and adapting regulations are vital to navigate this technological evolution, ensuring ATCOs remain at the core of a safe and efficient air traffic management system. The future of ATC balances technological advancement with the indispensable human element, leveraging AI as a tool to enhance, not replace, the expertise of skilled ATCOs.

Read the full article: https://www.eurocontrol.int/article/digitalisation-and-ai-air-traffic-control-balancing-innovation-human-element

#AI #ArtificialIntelligence #AirTrafficControl #ATC #AviationSafety #AIinAviation #DigitalATC #Nexyad

St Germain en Laye, January 10th 2025.

The combination of large language models (LLMs), like GPT, with genetic algorithms (GAs) is an exciting and emerging area of generative AI. This hybrid approach blends the adaptability and exploration capabilities of genetic algorithms with the vast language understanding and generation power of LLMs. Such a synergy has the potential to open up new frontiers in various domains, from optimization problems to creative applications and beyond.

Key Trends and Research Directions

• Evolutionary Language Models
  • Genetic Algorithms for Fine-Tuning LLMs: Instead of traditional gradient-based fine-tuning, genetic algorithms (GAs) can be used to evolve and optimize the architecture of LLMs, their hyperparameters, or even fine-tuning strategies. The evolution process could generate novel solutions for specific tasks by evolving networks that fit specific needs, like improving the performance of a language model on a rare domain or task.
  • Neuroevolution of LLMs: Using evolutionary strategies, researchers can evolve neural networks that are better suited for specific applications of language models. For example, evolving architectures for natural language processing (NLP) tasks like sentiment analysis, text summarization, or machine translation.

• Generative Design and Creativity
  • Creative Content Generation: A hybrid LLM and GA could be used to generate new types of creative content (e.g., art, music, stories). Genetic algorithms can evolve new prompts or even combinations of existing generative models, iterating through many generations of content generation, leading to innovative and diverse outputs.
  • Procedural Content Generation (PCG): In game design, genetic algorithms can evolve new levels, characters, or entire game worlds. Coupled with LLMs, this can lead to automatically generated narratives, quests, and character dialogues that change based on player behavior.

• Optimization in AI Systems
  • Evolving Hyperparameters and Architectures: By applying genetic algorithms, LLMs can be optimized to better perform on specific tasks by selecting and evolving the best performing architectures or hyperparameters. Genetic algorithms could also help in automatically discovering new optimization strategies for LLMs.
  • Automated ML (AutoML): GAs could play a role in automating machine learning workflows, such as finding the most optimal feature sets, training datasets, and model architectures by evolving various configurations to maximize performance.

• Human-Machine Interaction
  • AI Co-Creation Systems: Using genetic algorithms, generative AI systems can be designed to evolve user interactions based on feedback loops. This can result in AI systems that continuously adapt their behaviors to meet users’ needs in creative, productive, and personalized ways.
  • Language and Personality Evolution: Through genetic algorithms, LLMs could evolve more effective dialogue systems with better contextual understanding, adapting to various user preferences, tones, and conversational styles. AI-driven agents could simulate personalized interaction styles over time, based on user feedback.

• Robotics and Autonomous Systems
  • Evolving Communication Protocols: Combining LLMs and genetic algorithms, robots could evolve new communication strategies and protocols that enhance collaboration between machines or between human-robot teams.
  • Autonomous Decision Making: LLMs with GAs could help robots optimize complex decision-making processes, where evolutionary processes generate strategies for robotic navigation, interaction with environments, or coordination tasks.

• Synthetic Biology and Drug Design
  • Evolving Proteins or Genes Using LLMs: LLMs trained on protein structure data or genetic sequences can be integrated with genetic algorithms to evolve novel proteins or genes for drug development or synthetic biology.
  • AI-driven Biocomputing: Genetic algorithms can be employed alongside language models in synthetic biology, enabling the creation of new computational models based on genetic and protein structure data.

Research Projects and Applications

• Meta-Learning and Evolutionary Strategies for AI

1. • Several studies are exploring meta-learning frameworks where evolutionary algorithms guide the adaptation of LLMs in complex environments. This could be used in areas like few-shot learning or transfer learning, where LLMs need to rapidly adapt to new tasks with minimal data.
   • Example: Projects at institutions like OpenAI, DeepMind, and academic labs are experimenting with combining genetic algorithms and reinforcement learning for meta-learning, where evolution drives model optimization over multiple generations.

• Artificial Creativity and AI Art

1. • AI-generated Art Evolution: Projects like Artbreeder use evolutionary algorithms to combine and modify images, allowing users to explore creative possibilities. Such approaches can integrate language models to evolve visual narratives or storytelling through generative art.
   • AI Music Evolution: Genetic algorithms applied to LLMs in the music generation space have led to systems that produce musical compositions, evolving over time based on user feedback or pre-defined goals like genre, mood, or tempo.

• Optimizing Language Models for Specialized Domains

1. • Scientific Research: A promising application is evolving language models to specialize in specific scientific fields. By applying genetic algorithms to fine-tune large models on domain-specific data, researchers can develop specialized models that handle niche topics (e.g., genomics, climate science).
   • Example: Projects at institutions like MIT or Stanford are evolving LLMs using genetic algorithms to advance specific applications such as medical diagnosis, legal text analysis, or academic research.

• Hybrid Learning Systems

1. • Combining LLMs with evolutionary strategies can lead to hybrid systems that outperform traditional deep learning methods. These hybrid systems can optimize learning, selecting from various training methodologies and solutions to achieve better results.
   • Cognitive Architectures: Genetic algorithms can be used to evolve cognitive models of reasoning, which are then combined with LLMs to create advanced intelligent systems capable of general reasoning.

Emerging Applications

• Automated Scientific Discovery: Evolving AI models to assist in scientific discovery, where LLMs can be combined with GAs to generate hypotheses, test them, and evolve successful theories. For instance, evolving models that predict molecular interactions for drug discovery.
• AI-Driven Personalization: Genetic algorithms could personalize language models to adapt to individual users’ needs over time, tailoring interactions and content generation based on personal data.
• Robotic Systems with Evolving Dialogues: Imagine a system where robots or AI-driven assistants evolve their communication strategies and adapt to users’ preferences or behavior patterns over time, making them more effective over long-term interactions.

In conclusion, the intersection of large language models and genetic algorithms is paving the way for innovative research and applications. From optimization and fine-tuning of language models to creative generative applications and personalized AI systems, the potential for this hybrid approach is vast. Research is still in its early stages, but it’s clear that combining these technologies could lead to AI systems that are more adaptable, creative, and capable of solving complex, real-world problems.

#LLMs #GeneticAlgorithms #GenerativeAI #Neuroevolution #AIoptimization #CreativeAI #HybridAI #EvolutionaryComputation #ArtificialIntelligence #AIresearch #AI #Nexyad

St Germain en Laye, January 9th 2025.

The future of AI, particularly the integration of knowledge-based systems (KBS) and deep learning (DL), promises to be a rich, transformative area, combining structured, human-readable knowledge with data-driven, pattern recognition techniques. This hybrid approach leverages the strengths of both paradigms, fostering more robust, explainable, and adaptable AI systems. Some key trends and projects shaping the future of this intersection include:

Neural-Symbolic AI

Neural-symbolic systems are designed to integrate deep learning’s pattern recognition capabilities with knowledge-based reasoning. These systems aim to combine the representational power of symbolic logic (as seen in KBS) with the learning flexibility of neural networks. Key areas of development include:

• Knowledge-Enhanced Models: Deep learning models, such as transformers, being enhanced with explicit knowledge representations, like ontologies or relational databases, for better understanding and reasoning.
• Explainability & Interpretability: By adding symbolic structures to neural networks, these systems offer clearer reasoning paths, making AI decisions more understandable and transparent.
• Projects:
• • DeepMind’s Neuro-Symbolic AI: DeepMind has worked on combining neural networks with symbolic reasoning to build AI that can reason about the world in a more human-like manner.
  • IBM’s Neural-Symbolic Learning and Reasoning: IBM is exploring combining deep learning and knowledge graphs to create systems that can learn and reason in complex environments.

Knowledge Graphs and Deep Learning

Knowledge graphs (KGs) are a form of knowledge-based system where data is represented as entities and the relationships between them. Deep learning models are increasingly being used to enhance KGs for tasks like semantic search, natural language understanding, and reasoning.

• Hybrid Systems: Using deep learning to process unstructured data (text, images, etc.) and knowledge graphs to provide contextual understanding, AI systems can better simulate human knowledge and reasoning.
• Trends:
  • Graph Neural Networks (GNNs): GNNs are gaining traction as a way to incorporate graph-based data structures (such as KGs) into deep learning models, enabling systems to make more informed predictions and reasoning.
  • Knowledge-Augmented NLP Models: Models like GPT-3 and BERT can integrate KGs to improve comprehension, context understanding, and decision-making in natural language processing tasks.
• Projects:
  • Google’s Knowledge Graph: Google is heavily investing in combining deep learning with its knowledge graph for more accurate search results and enhanced AI systems.
  • Facebook’s Knowledge Graph-Based AI: Facebook is integrating deep learning with knowledge graphs to improve its recommendation systems and knowledge retrieval.

Cognitive Architectures

Cognitive architectures attempt to simulate the human brain’s way of processing information by integrating various AI components, including KBS and deep learning. These architectures aim to model reasoning, learning, memory, and perception in a unified system.

• Trends:
  • Memory-Augmented Neural Networks (MANNs): These models integrate memory networks with deep learning to help machines retain and recall previous experiences or data, enhancing their decision-making and reasoning abilities.
  • Cognitive AI for Autonomous Systems: This trend is driving the development of cognitive agents capable of interacting with complex environments in a human-like manner by leveraging both structured knowledge and experience.
• Projects:
  • ACT-R: A cognitive architecture designed for modeling human cognition, which combines symbolic and sub-symbolic processing. There is growing interest in applying this model to AI.
  • DARPA’s AI Exploration on Cognitive Architectures: The U.S. Department of Defense’s DARPA is investing in developing cognitive systems that combine symbolic reasoning with deep learning for military applications.

Multimodal AI

Multimodal AI is an emerging field that combines multiple types of data (text, images, video, sound, etc.) to create richer, more nuanced AI models. The fusion of KBS and deep learning in multimodal systems allows for advanced reasoning across different types of information.

• Trends:
  • Vision-Language Models: Models like CLIP (Contrastive Language-Image Pretraining) and DALL-E from OpenAI combine vision and language for better contextual understanding, using deep learning and knowledge-based cues to improve image generation or text-to-image understanding.
  • Cross-Modal Retrieval: AI systems will be able to retrieve and reason across text, audio, and visual data using both deep learning and knowledge graphs for richer interaction and analysis.
• Projects:
  • OpenAI’s CLIP and DALL-E: Both are leading multimodal systems using deep learning and knowledge concepts to generate images from text descriptions or perform cross-modal understanding tasks.
  • Google’s Multimodal AI (Pathways): Google’s Pathways model aims to combine various AI capabilities, like language understanding and visual processing, in a unified system.

AI for Scientific Discovery

The hybrid approach of KBS and deep learning is being increasingly applied in domains like healthcare, drug discovery, and materials science, where vast amounts of structured and unstructured data need to be processed and understood.

• Trends:
  • AI-Assisted Drug Discovery: Combining knowledge about biological processes with deep learning-based prediction models to accelerate the design of new drugs.
  • Simulations and Forecasting: Knowledge-based systems are being integrated with machine learning to model complex systems (like climate change, economic trends, or epidemic outbreaks) with better accuracy.
• Projects:
  • Insilico Medicine: This company is applying AI and deep learning techniques to drug discovery, leveraging structured biological data and deep neural networks for predictions.
  • DeepMind’s AlphaFold: Using deep learning to predict protein folding, which represents a knowledge-based integration between biology and computational methods.

Ethics, Governance, and Fairness in AI

With the increased complexity of AI systems combining deep learning and knowledge-based approaches, there is a growing need for ethical frameworks and governance mechanisms to ensure fairness, transparency, and accountability.

• Trends:
  • Explainable AI (XAI): There is a growing interest in developing explainable AI models that combine deep learning’s predictive power with knowledge-based systems’ transparency.
  • Fairness and Bias Mitigation: Research into ensuring that AI models built from both structured and unstructured knowledge are fair, unbiased, and ethical.
• Projects:
  • Microsoft’s Fairness and Transparency in AI: Microsoft is working on techniques to ensure that AI models built with deep learning and knowledge graphs adhere to ethical standards.
  • AI Ethics and Governance: Governments and organizations like the European Union are setting up frameworks for AI development that emphasize transparency, fairness, and accountability.

In conclusion, the future of AI, where knowledge-based systems and deep learning converge, is poised to enhance many aspects of our world, from healthcare to autonomous systems to scientific research. With the hybridization of these fields, AI is becoming more adaptable, explainable, and intelligent, creating more human-like systems that combine the strengths of both structured knowledge and deep learning. Key trends to follow include neural-symbolic integration, multimodal AI, and ethical considerations, with projects in drug discovery, cognitive architectures, and AI fairness paving the way for the next wave of AI innovation.

#ArtificialIntelligence #DeepLearning #KnowledgeBasedSystems #NeuralSymbolicAI #CognitiveArchitecture #AIForScience #ExplainableAI #MultimodalAI #EthicalAI #Nexyad #KnowledgeGraphs

St Germain en Laye, January 8th 2025.

The use of a prudence metric calculated in real time, like the prudence metric proposed by NEXYAD, offers several strategic advantages for a fleet, particularly in the areas of road safety and lower operating costs. Below is a breakdown of these benefits:

Improved Road Safety

• Risk Management: A real-time prudence metric can assess driving behavior and identify potential risks as they happen. For example, it could measure factors like speed, harsh braking, acceleration, or cornering. Fleet managers can receive alerts about risky driving behaviors that might lead to accidents.
• Preventive Measures: By monitoring the metric, drivers can be prompted to adjust their behavior before a dangerous situation occurs. This can help prevent accidents, injuries, and fatalities, contributing to a safer driving environment.
• Driver Feedback: Real-time insights allow for immediate feedback to drivers, promoting safer driving habits and reinforcing positive behaviors, which can reduce the likelihood of incidents over time.
• Accident Cost Reduction: Reducing road accidents decreases costs associated with vehicle repairs, legal liabilities, insurance claims, and downtime.

Lower Operating Costs

• Fuel Efficiency: A prudence metric can detect fuel-inefficient behaviors (e.g., speeding, rapid acceleration) and encourage more fuel-efficient driving techniques. This can lead to reduced fuel consumption and, consequently, lower fuel costs.
• Maintenance Savings: Monitoring driving behaviors in real time also helps to reduce wear and tear on the vehicles. Harsh braking and rapid acceleration lead to quicker degradation of engine components and braking systems, leading to more frequent maintenance and higher repair costs. By encouraging more prudent driving, fleets can extend vehicle life and reduce maintenance expenses.
• Optimized Routing: The prudence metric can be integrated with other systems like GPS tracking, providing real-time insights into traffic conditions, route changes, and optimal paths. Efficient route planning helps to avoid unnecessary detours or congestion, reducing both fuel consumption and travel time.
• Insurance Savings: Fleets that use real-time metrics and demonstrate improved safety records may qualify for discounts on insurance premiums. Insurance companies often provide lower rates to fleets that use advanced technologies to track driver behavior and reduce risk.
• Driver Retention and Productivity: Encouraging good driving habits can lead to a more motivated and satisfied driver workforce, which in turn reduces turnover and associated hiring/training costs. Additionally, safer and more efficient driving results in fewer breakdowns, delays, and service interruptions.

Enhanced Fleet Management

• Data-Driven Decisions: Real-time prudence metrics give fleet managers up-to-date insights into the overall performance of the fleet. They can identify trends, weaknesses, or areas for improvement, making it easier to implement targeted strategies for efficiency and safety.
• Compliance and Reporting: For fleets operating under strict regulations (e.g., hours of service, speed limits), real-time tracking of driving behaviors can help ensure compliance with local laws and industry standards. This reduces the risk of fines or legal issues.

By using a real-time prudence metric, fleets can proactively manage road safety, reduce operating costs, and make more informed decisions. This data-driven approach enhances overall fleet performance and ensures a balance between safety, efficiency, and cost-effectiveness.

#Nexyad #SafetyNex #FleetManagement #RoadSafety #PrudenceMetric #RiskManagement #FuelEfficiency #DriverFeedback #CostReduction #DataDriven #VehicleMaintenance #InsuranceSavings

St Germain en Laye, January 7th 2025.

Why “Living Intelligence” Is the Next Big Thing

an article by Amy Webb in Harvard Business Review

« AI is just one of the three revolutionary technologies transforming the business landscape. The other two—advanced sensors and biotechnology—are less visible but no less important, and have been quietly advancing. Soon, the convergence of these three technologies will underpin a new reality that will shape the future decisions of every leader across all sectors. »

The author thinks that AI represents just one aspect of a broad technological transformation happening today. Companies that overlook the significance of other emerging technologies risk falling behind. Advanced sensors and biotechnology might be less prominent but are equally crucial and progressing steadily. The integration of these three technologies will soon form a new landscape, influencing the strategic decisions of leaders across various sectors. This emerging landscape, known as “living intelligence,” involves systems capable of sensing, learning, adapting, and evolving—powered by the synergy of artificial intelligence, advanced sensors, and biotechnology. Living intelligence is poised to fuel an accelerated cycle of innovation, reshaping industries, and spawning entirely new markets. Leaders who concentrate only on AI, ignoring its interplay with the other two technologies, may miss an impending wave of disruption.

Why “Living Intelligence” Is the Next Big Thing

#LivingIntelligence #AIInnovation #TechConvergence #FutureOfHealthcare #AI #DigitalTransformation #SensorTechnology #BiotechnologyAdvancements #DataDrivenDecisions #EmergingTechnologies #OrganizationalStrategy #HealthcareInnovation #TechDisruption #Nexyad

St Germain en Laye, January 6th 2025.

Generative AI is revolutionizing the landscape of research by enabling unprecedented levels of automation and innovation, and facilitating major breakthroughs across all research fields. To equip researchers with the necessary skills to transform their research with generative AI, the Michigan Institute for Data Science and the Michigan AI Laboratory jointly offered a series of tutorials in the Fall of 2023, sessions consisting of lectures and hands-on demonstrations by U-M experts to train researchers in using generative AI for improving research workflows, coding, building custom models and more.

« Using Generative AI to Assist in Coding in Data Science Research » – Dr. Sindhu Kutty, Lecturer IV in Electrical Engineering and Computer Science, College of Engineering, University of Michigan

St Germain en Laye, January 2nd 2025.

St Germain en Laye, December 31st 2024.

The first time Nexyad heard about teleoperation for autonomous vehicles was in 2017 during a BMW Startup Garage event, where Nexyad was invited after winning a tech date contest. The concept involved an operations center in a large city, with one person monitoring ten to fifteen autonomous vehicles simultaneously, able to take remote control if a problem arose.

In France, the MILLA Group offers a similar teleoperation service to its customers. Their software platform integrates all operational assistance and passenger information functions, as well as remote management of autonomous shuttles.

In the U.S., Tesla is gearing up to launch a robotaxi service and is expanding its teleoperations team to support this effort. A recent job listing for a software engineer suggests that Tesla is developing a teleoperations system that will allow human operators to remotely control its robotaxis and humanoid robots. This marks a strategic shift away from Tesla’s previous focus on achieving full autonomy without human intervention, as emphasized by CEO Elon Musk.
Teleoperations are seen as vital in the autonomous vehicle industry, helping to manage complex scenarios like construction zones and accidents. While Tesla has previously used teleoperations for its Optimus robots, operating robotaxis will have different requirements, demanding advanced user interfaces and reliable communication systems.

Tesla recently unveiled its prototype robotaxi, the Cybercab, which is expected to enter production around 2026 or 2027. Musk has also mentioned plans to launch a self-driving ride-hailing service in California and Texas by 2025, with testing already in progress. It’s still unclear whether the new teleoperations team will assist only robotaxis or also Tesla vehicles currently on the roads.
Overall, the formation of the teleoperations team and system indicates Tesla’s growing commitment to its robotaxi initiative, even as Musk’s ambitious timelines have led to skepticism due to previous overpromises.

Nexyad Driving prudence Measurement can be precious in AD systems of Autonomous Vehicle as prudence feedback of closed loop on control, instead of classical open chain where every elements (sensors, perception, planning) has to be perfect to make control efficient and secure.
But in case robotaxis companies don’t want to touch control onboard, they can have at least the information of prudence/risk in their teleoperation room.

#Nexyad #BMW #Milla #Tesla #ElonMusk #Robotaxi #AutonomousVehicle #SafetyNex #DrivingPrudenceMeasurement #CyberCab #AutonomousShuttle #AI

St Germain en Laye, December 30th 2024.

Nexyad SafetyNex hybrid AI is a software, based on road safety knowledge, which allows to ANTICIPATE upcoming EVENTS like curves, intersections, speed limits, pedestrian crossings, school zones, weather conditions, accidents, road works, traffic congestion, merging traffic and even more other singularities.

It adjusts the vehicle’s speed BEFORE the driver needs to react, smoothing out the driving experience and potentially improving fuel efficiency.
SafetyNex can use data from maps (SD/HD), objects detections sensors and information from telecom.
It is the perfect tool to build PREDICTIVE system (ACC / Automated Driving).

It makes a big difference with systems that focuses on reacting to immediate threats, using emergency braking when it’s time to avoid collision, the Preventive ADAS.

At Nexyad, we use the glass on the table to illustrate this difference:
. anticipation is keeping out the glass of the edge for serenity
. reaction needs good reflex and generates anxiety

See our page https://nexyad.net/Automotive-Transportation/prudence-based-automated-driving/

#Nexyad #SafetyNex #AI #ADAS #PredictiveACC #Anticipation #AutomatedDriving #AutonomousVehicle

St Germain en Laye, December 23th 2024.

The autonomous vehicle is supposed to drastically reduce accidents and save lives. Road accidents in the West and even more so in the rest of the world are a major cause of mortality. We can recall the number of victims per year which amounts to 1.2 million people today. Much more than the victims of wars in Europe, the Middle East and all conflicts combined.

Automotive engineers know how to build safe cars, which brake very hard, which hold the road well, which do not explode while moving and which are structurally designed to protect the occupants in the event of an impact. On the other hand, they do not know much about road safety. This discipline is extremely complex. Accidents are fortunately rare events if we consider the number of miles traveled where nothing happens.

Road safety specialists are generally civil servants. They work on infrastructure and have an understanding of how accidents happen, because they receive and analyze information by road police, firefighters, etc.

Over the past 20 years, Nexyad has invested in 12 funded scientific research programs on road safety. It involved collaboration among hundreds experts, professional drivers, police of the road, and insurers from 19 countries. The first time by chance, then to capitalize on this knowledge. We met experts from 19 countries, we compared their opinions, defined a common vocabulary and finally synthesized all this knowledge. We extracted a corpus of rules of prudent driving. To make these rules simply usable, we built a metric of prudent driving. The metric is scaled from 0 to 100%. We designed it so that (the last decile) of the scale, from 90% to 100%, represents what road safety experts consider the 95% most dangerous or accident-prone driving situations.

Predit ARCOS 2004 – Predit SARI 2006 – MERIT 2006 – SURVIE 2009 – CENTRALE OO 2011 – CASA 2014 – SERA 2014  AWARE 2014 – RASSUR 79 2015 – SEMACOR 2015 – SEMACOR 2 2017 – BIKER ANGEL 2020

Today, we use this metric to measure in real time and on board  driving of both humans and robotized drivers. When we detect lack of prudence, we can alert humans and we give feedback to control of autonomous driving with enough anticipation allowing time to slow down or/and change trajectory.

See our website pages:

Autonomous Cars & Autonomous Trucks driven by Prudence

BYOD Solutions for Fleets: Bring Your Own Device Solutions

#Nexyad #AI #ArtificialIntelligence #BYOD #AutonomousDriving #AV #ADAS #SafetyNex #DrivingPrudenceMetric #RoadSafety #Science #AccidentsKnowledge