Articles tagged with Machine Learning
Researchers from MIPT and MSU created a computer model predicting agrochemical activity using machine learning and Kohonen self-organizing maps. The model showed high predictive power, accurately classifying molecules into pesticides or plant growth regulators with 87% accuracy.
A new robotically driven experimentation system has been developed to determine the effects of a large number of drugs on many proteins, reducing the number of necessary experiments by 70%. The model uses an active learning approach to identify patterns and make predictions about unmeasured experiments with high accuracy.
Researchers at University of California, Riverside developed an energy management system that improves plug-in hybrid efficiency by 12 percent. The system uses machine learning and real-time data to optimize energy consumption and reduce greenhouse gas emissions.
Researchers will use insights from neural circuitry and learning methods to create more human-like computer vision and machine learning algorithms. The CMU-led team aims to unlock the brain's secret algorithms in learning and inference with a massive database of neural activity.
Researchers at Harvard are using a $28 million grant to study the brain's visual cortex in unprecedented detail and map its connections. The goal is to inspire better computer algorithms for learning and pattern recognition, enabling computers to outperform humans in recognizing patterns from limited data inputs.
A team of researchers is exploring the use of the NAO robot in a new approach to pediatric rehabilitation based on social interaction between robots and humans. The robot can read moods, recognize family members, and learn preferences, providing personalized interventions for children with motor disabilities.
InSilico Medicine presents recent advances in applying signaling pathway activation analysis and deep learning to drug discovery and age-related diseases. The company's mission is to extend healthy human longevity through faster and more effective diagnostics and cures.
Researchers at Carnegie Mellon University have developed a high-throughput, machine-learning tool to analyze synaptic density in the brain. This allows them to identify synapses from an entire cortical region and gain insights into how synaptic properties change during development and learning.
Researchers found that humans can recognize objects even when only a small portion is visible, and validated an algorithm to explain human learning. The algorithm can be used for machine learning, data analysis, and computer vision to improve performance.
Researchers developed a Bayesian Program Learning framework that captures human learning abilities, allowing computers to recognize and generate new visual concepts. The algorithm achieved impressive results in visual Turing tests, with only 25% of judges performing better than chance.
Researchers developed a computer model called Bayesian program learning framework (BPL) that captures humans' unique ability to learn new concepts from a single example. The BPL model achieved human-level performance on challenging concept learning tasks, outperforming recent deep learning approaches.
Researchers at the University of Washington have created a new probabilistic model that allows robots to learn new skills by watching people and imitating them. The team combined child development research with machine learning approaches, inspired by infants' ability to infer adult intentions through self-exploration.
The study aims to develop machine learning models that mimic the toddlers' ability to recognize objects, leveraging 500 hours of video and 54 million images from over 100 children. The research could lead to more sophisticated digital object-recognition technology.
Researchers at UMass Amherst aim to develop a reliable, predictive computational framework for designing better-performing materials with reduced development costs. The new approach will address challenges in handling complex systems with millions of variables, rare events, and multi-scale features.
Researchers at the University of Wisconsin-Madison are developing a new approach called machine teaching, which uses sophisticated mathematics to model human learners and devise the best possible lessons. This method has immense potential to impact education by providing optimal, personalized lessons for students in various fields.
Researchers developed a tool enabling art directors to control computer programs using verbal descriptors like 'silky' and 'wrinkly'. The system improved the process of creating garments with desired properties, reducing laborious tweaking of technical parameters.
Researchers at University College London developed a novel AI approach to predict fine wine prices more accurately, outperforming traditional methods by 15% on average. The new method uses machine learning to learn relevant information from data and improve predictive accuracy.
DeepBind uses deep learning to analyze protein-DNA/RNA binding and detect mutations that can disrupt cellular processes. The tool provides new information on disruptions in mutations tied to cancers, haemophilia, and familial hypercholesterolemia.
Recent progress in AI has driven significant advancements in machine learning, language processing, and human-like computer programs. Researchers discuss potential implications for humankind, including challenges to privacy, equality, and the public good.
Researchers developed a dropout-prediction model that uses data from one course offering to predict stopout in the next. The model achieved fairly accurate predictions and showed promise, particularly when incorporating additional variables like weekend study habits. Ongoing work aims to refine the model for improved accuracy.
A new study uses machine learning to pinpoint rodent species effective disease reservoirs and identify geographic hotspots vulnerable to emerging diseases. The research provides a basis for targeted surveillance efforts, highlighting the importance of collaboration with experts on the ground.
Researchers at MIT have developed a probabilistic programming language called Picture that can solve computer-vision tasks using short programs. The new system, which is competitive with conventional systems, has been shown to improve error rates on certain tasks, such as human pose estimation.
Disney Research has developed a robotic camera system that can learn from human operators to better frame shots of a basketball game. The system uses machine learning algorithms to recognize the relationship between player locations and corresponding camera configurations.
A new automated method developed by Disney Research uses AI to select and order photos in a way that makes narrative sense, telling a compelling story. The system learns principles of selecting and ordering photos from large collections and can customize the process for individual preferences.
A new system developed at MIT enables pattern-recognition systems to distill what they learn into simple examples, which humans can use to make better decisions. In experiments, human subjects using the system outperformed those using a similar algorithm by over 20 percent.
Researchers at Carnegie Mellon University used fMRI scans to analyze brain activity while eight people read a chapter of Harry Potter. The result was the first integrated computational model of reading, identifying which parts of the brain are responsible for parsing sentences and determining word meaning.
A team of scientists developed a mathematical model that explains how the brain maintains stability during learning, resolving a decades-old paradox. The model suggests that fast and slow changes in neuronal networks work together to achieve sensitivity and stability.
Quantum computing enables robots to learn and adapt faster, with a significant speedup in response times. This breakthrough has implications for machine learning, climate modeling, and internet search engines, leading towards a more ambitious objective of creating intelligent and creative robots.
Researchers developed AI software that predicts player goals in video games with 62.3% accuracy, outperforming previous technology. The software uses deep learning to analyze large collections of game data and improve its accuracy over time.
Researchers used brain-computer interfaces and machine learning to study neural patterns in monkey brains as they learned to move a computer cursor. The study found that learning was easier when nerve cells rearranged existing patterns of activity, rather than generating new ones.
A 1996 research paper by Michael Pazzani and colleagues has been selected as the most influential from The Thirteenth National Conference on Artificial Intelligence. The paper introduced a system to personalize internet content based on user profiles, which has since become a common application of artificial intelligence.
UCI researchers develop computing techniques that utilize deep learning to analyze data from particle accelerators, increasing the detection rate of rare particles by 8%. The methods could aid in the hunt for fundamental open questions about matter, gravity, and the origin of the universe.
The new Birdsnap app, developed by Columbia Engineering researchers, can identify 500 common North American bird species using computer vision and machine learning techniques. It offers users various ways to organize species and even annotates images with distinctive parts for easy identification.
Researchers have demonstrated that powerful probabilistic reasoning algorithms can be implemented using chemical reactions, enabling the creation of intelligent machines at tiny scales. This breakthrough could lead to the development of 'smart drugs' that can automatically detect, diagnose, and treat diseases.
A Cornell robot has been programmed to anticipate human actions, enabling it to step in and offer a helping hand. The robot uses a Microsoft Kinect camera and a database of 3D videos to identify activities and predict future actions.
Professor Murat Dundar will use the award to refine machine learning models for more accurate data classification and adaptation. His research aims to improve bio-detection, medical monitoring, and data sorting technologies.
Arthur Szlam, assistant professor at City College of New York, has been awarded the Sloan Research Fellowship for his innovative mathematics in machine learning and computer vision. He will receive $50,000 to further his research on computer vision, allowing computers to learn to distinguish and categorize objects in images.
Researchers Todd Gureckis and Douglas Markant examine the benefits of self-directed learning from a cognitive and computational perspective. They argue that this approach optimizes educational experiences by focusing on useful information and exposing learners to new sources. By understanding these processes, researchers can develop as...
AEMASE is a cognitive software application that updates its knowledge of experts' performance in real-time, automatically evaluating student performance and reducing overall training costs. The system will be used to train Navy personnel on various aircraft, including the H-60 helicopter and E-2C Hawkeye aircraft.
Researchers at Georgia Tech identified key question types that facilitate human-robot learning, including feature queries, which were preferred by both human volunteers and robot learners. The study aims to improve the teaching of robots by understanding human learning mechanisms and developing more effective active learning strategies.
Researchers at the University of Bristol used musical features and machine learning algorithms to predict song hits in the UK singles chart. They found that danceability increased in popularity from the late 1970s and that slower styles, such as ballads, were more likely to become hits in the 1980s.
Researchers at Oregon State University have developed a new system that combines computer vision, machine learning, and automated planning to improve operations in various industries. The system is based on analyzing football plays and can be applied to tasks such as factory efficiency, airport operation, and nursing care.
Researchers have developed a computational tool that can determine whether faces are attractive, threatening or dominant with high accuracy. The tool uses machine learning techniques to analyze facial characteristics and was tested on a set of synthetic images, achieving accuracies of up to 96%.
Researchers at Harvard University's Neuromotor Control Lab found that motion-referenced learning, where the brain learns from actual movements rather than intended actions, can improve learning efficiency. This approach may lead to more effective neurological rehabilitation for individuals with stroke or other motor disorders.
Researchers at Tel Aviv University have developed an algorithm that enables computers to anticipate the future and make more efficient decisions. The algorithm, funded by Google, aims to minimize 'regret' in computer decision-making by analyzing variables and adapting to situations.
Researchers at EPFL developed a brain-computer interface that learns to recognize users' mental intentions, allowing for multitasking and reducing fatigue. The system uses statistical analysis and probability theory to distinguish between commands and enable users to control devices over longer periods.
Researchers created a machine-learning model called DiaTM that learns vernacular terms for health problems and symptoms, improving medical website performance. The system achieves a 25% improvement in nDCG, a scientific term referring to the relevance of information retrieval.
Researchers studied infant-mother interactions to develop a baby robot capable of learning social skills. They found that babies and mothers create patterns in their play, which becomes more stable with age.
A study found that nearly a quarter of the variability in achievement among video gamers can be predicted by measuring specific brain structures. Players with larger nucleus accumbens, caudate nucleus, and putamen performed better in training periods.
A Carnegie Mellon researcher notes that data-mining techniques are increasingly being applied to personal activities and movements, raising concerns about privacy. Technical means can help limit these risks, but a public discussion about data collection, ownership, and privacy is also necessary.
Researchers develop algorithms that allow end users to ask computers why they made mistakes, read their responses, and explain why those were errors. This 'meaningful' interaction enables computers to customize themselves to users and perform better in the future.
The new science of learning emphasizes computational, social, and brain-based approaches to understanding human learning. Key findings include the importance of machine learning, social interaction, and empathy in learning, which are now being applied to develop personalized teaching tools.
Research highlights three principles: learning is computational, social interaction underpins early learning, and brain circuits linking perception and action support learning across the life span. Social interaction with humans, particularly parents or tutors, plays a crucial role in early learning.
Researchers at UC San Diego used machine learning to empower their Einstein robot to learn realistic facial expressions, improving the process of teaching robots to make lifelike faces. The team discovered that the model learned to automatically compensate for missing servos and can now make facial expressions it had never encountered.
The Pittsburgh Science of Learning Center (PSLC) will continue its research on how people learn with a renewed $25 million NSF grant. The PSLC conducts experiments in over 50 classrooms across the US to understand learning styles and habits.
Researchers aim to develop new computational models of visual system learning and uncover mechanisms that explain the learning process in neural circuits. The project seeks to understand the role of feedback connections in the visual cortex during learning.
The University of Washington is leading a $6.25M project to develop an electronic Sherlock Holmes system for complex data analysis in the military. The system will integrate various types of sensor data to predict behavior and make decisions, addressing the challenge of handling high degrees of complexity and uncertainty.
The NSF has launched three Science of Learning Centers to study the foundations of learning across various situations, from cellular to complex processes engaging different brain areas. The centers will support interdisciplinary research and develop new methods for improving human learning and developing intelligent machines.
A NASA grant is supporting a two-year study at Florida Tech to develop algorithms that can learn from historical data and detect potential problems with a space shuttle component. The research aims to improve the efficiency of monitoring systems, reducing the time and effort required to extract knowledge from experts.