From AI perfume to AI beauty algorithm, can the aesthetic optimal solution be found?
The information that the artificial intelligence system needs to perceive mainly comes from its sense of sight, hearing, and touch, and the sense of smell is also a very valuable direction. At present, researchers have been studying odor sensors and their applications in environmental protection and other fields.
During the CIIE, in addition to "hard-core" smart technologies such as bionic robots and automated production lines, AI beauty consultants, virtual makeup artists, and customized skin care products driven by artificial intelligence also attracted the attention of many people.
In recent years, artificial intelligence, which shines in the fields of image, video and voice recognition, has gradually penetrated into the beauty industry. The market has shown that perfumes formulated by artificial intelligence have received good feedback from consumers.
"Preparation of perfume is a complex systematic project involving raw materials, processes, users, markets and many other factors. The workload of searching for the right solution in the vast parameter space is huge. It is difficult to exhaust all feasible solutions only by relying on manual experience. You may miss the best solution that is most popular with the market and users." Sun Zhenan, a researcher at the Institute of Automation, Chinese Academy of Sciences, said in an interview with a reporter from Science and Technology Daily.
AI fragrance does not rely on fragrance to judge the order of spices
"The information that the artificial intelligence system needs to perceive mainly comes from its vision, hearing, and touch, and smell is also a very valuable direction. At present, researchers have been studying odor sensors and their applications in environmental protection and other fields." Sun Zhenan said.
Prior to this, IBM had created an artificial intelligence perfume apprentice-Philyra.
Philyra is a creative artificial intelligence that can study existing perfume formulas and compare their ingredients to develop new perfume formulas.
Philyra can also get the popularity of perfumes in different genders, ages, and best-selling places. After combining with big data algorithms, Philyra will output a new perfume formula that can achieve good results in the intended target population.
Sun Zhenan explained that unlike perfumers who take decades to train their sense of smell, artificial intelligence does not rely on smell to make perfume.
AI perfumery uses advanced machine learning algorithms to analyze and learn information such as perfume formulas, raw materials, historical sales data and industry trends to predict human preferences, thereby creating new perfume formulas for the target population.
Perfumer David Appel conducted an experiment with Philyra: A perfume produced entirely by Philyra was A type; B was mainly made by Philyra, and the perfumer served as an auxiliary modification; C was modified by a perfumer. The incense master is the lead and Philyra is the support.
The three perfumes were tested after they were completed. The test results showed that the vast majority of people chose the A perfume created by artificial intelligence.
"Compared with artificial fragrances, AI fragrances do not rely on fragrances to determine the order of fragrances, but customize the blending method after analysis by deep learning algorithms. This is a method different from traditional fragrance thinking; AI perfumes will not be affected The influence of personal preference, experience, and cultural prejudice is to digitally quantify the sense of smell and explore more creative ingredient combinations and formula formulas; AI perfume can quickly calculate the most popular target customers with its powerful computing power The perfume formula shortens the perfume research and development cycle.” Sun Zhenan pointed out.
It is also used in scenarios such as smart medical beauty and hair design
According to Euromonitor Consulting’s forecast, my country’s cosmetics industry will maintain a compound growth rate of 8.3% from 2019 to 2023.
Faced with the continued growth of the cosmetics consumer market and the popularization of scientific skin care and functional skin care concepts, major beauty and skin care businesses are also actively using cutting-edge technologies such as artificial intelligence to launch customized products that are more in line with consumers’ personal needs and bring more Comprehensive product experience.
Sun Zhenan said that in addition to making perfume, artificial intelligence is also applied to many scenarios such as smart medical beauty, hair design, virtual makeup, customized skin care products, and skin diagnosis.
Researcher Sun Zhenan and Heran research team of the Institute of Automation of the Chinese Academy of Sciences have explored the application of facial image editing and beauty, medical beauty, and entertainment.
For example, they proposed a face age transformation technology based on wavelet domain at the 2019 IEEE International Computer and Pattern Recognition Conference, which can predict the change of face with age, and this technology can also be applied to personalized recommendation of skin care products; They proposed a face attribute editing technology based on sample importance sampling in 2020, which can edit attributes such as face skin color, hair color, adding or removing glasses, and can be applied to makeup design in the beauty salon industry. Users provide diversified choices and references; a face composition editing technology based on reference images proposed in 2020 can manipulate the facial features of the face, and assist in medical cosmetology, through preoperative prediction of plastic surgery and trauma repair surgery Doctors and patients will give more guidance and develop more scientific beauty programs.
"The application technology of artificial intelligence in the beauty and medical industry has saved labor costs for enterprises, improved the efficiency of customer service, and brought a convenient and comprehensive user experience for consumers to purchase products." Sun Zhenan said.
Sun Zhenan pointed out that from the current application, the personalized customized skin care system allows each user to have an exclusive skin care beautician, using face recognition and face analysis technology to analyze the photos provided by the user and evaluate their skin condition, and then The recommendation algorithm is used to select the most suitable product from the existing products for recommendation; the facial beauty evaluation system provides more valuable methods and tools for medical cosmetology, which can assist hospitals in cosmetic surgery, using the detected data and standard beauty The model is compared to provide the best solution for the improvement of each part; the AR smart makeup try uses face recognition and virtual makeup rendering algorithms to achieve the makeup effect of different brands and different colors on the face, reducing the cost of makeup. The beauty retail terminal has been expanded to meet the needs of fast fashion.
For aesthetic standards, humans and machines will influence each other
For the cognition of beauty, the benevolent sees the benevolent and the wise sees the wisdom. It is difficult to quantify and form a standard. However, "the love of beauty is shared by everyone." There are rules to follow in the aesthetic concept of the public. Everyone’s approval.
"Artificial intelligence's understanding of beauty requires big data training. At present, the definition and appearance of beautiful faces by artificial intelligence cannot be separated from human supervised data labeling, because artificial intelligence has no understanding of beauty itself. It has a grasp of beauty. It depends on the face data it learns and the artificially defined and quantified beauty label data set." Sun Zhenan said.
David Appel believes that artificial intelligence is not a threat but an object that requires cooperation. Machines have no excess feelings and prejudices, which can significantly enhance human creativity and provide the most innovative algorithms.
Previously, the CARTO artificial intelligence fragrance system was produced in the well-known perfume company Givaudan. The system uses the scent illustrations collected by IBM. It is hoped that the fragrance can maximize the olfactory performance in the formula.
For another example, Shiseido's Optune personalized skin care system consists of the smartphone application Optune App and the dedicated machine Optune zero.
Optune App uses machine learning technology to analyze the photos taken by the user to show the user's skin condition (skin quality, pores, moisture content, etc.), and sends the data to Optune Zero via the cloud, which selects from existing products based on an algorithm The most suitable combination of essence and moisturizing products is produced directly from the machine.
"In order to eliminate artificial intelligence's preference for human aesthetics, the data used by the training algorithm should be as comprehensive and diversified as possible, comprehensively multi-dimensional and multi-angle standards for learning. However, it does not rule out whether autonomous evolutionary intelligence has any effect on beautiful people and things. The supervising discovery ability, for example, the machine may automatically search for photos of celebrities that have attracted the attention of the general public to learn from the aesthetic standards." Sun Zhenan said.
In Sun Zhenan's view, with the application of artificial intelligence in the beauty industry, humans and machines will affect each other's aesthetic standards.
"On the one hand, humans have personal preferences for the standard of beauty in a broad sense. When algorithm experts use a certain definition of beauty as a standard for the machine to learn, the machine tends to favor the aesthetic standard of this definition. On the other hand, When the machine recommends what it thinks of medical beauty solutions for people for a long time, the user aesthetic system of the artificial intelligence system will also be affected."