Optimization techniques based on ant behavior tend to focus on the way these insects find the shortest path, but scientists at the CSIC Artificial Intelligence Institute have also studied how certain species can mark those that do not need to pass. The results can be used to solve logistics or drug search problems.
Well known ant They left a trail of pheromone and made the rest of the anthill follow the same route. The shorter the path to the nest, the more frequent the ants will pass: therefore, they accumulate more traces of pheromone and have a significant enhancement effect compared to other paths. This allows the community to find a short path.
this is”Swarm intelligenceThe collective behavior of animals such as ants, bees or termites has inspired the development of artificial intelligence. In fact, ACO technology (Ant colony optimization) It is based on the way ants find short distances and has been used in logistics, medical research or bioinformatics.
Now from Barcelona CSIC Institute of Artificial Intelligence (IIIA-CSIC) Inspired by ACO technology, improved ACO technology Pharaoh’s AntPharaoh oni), Can learn from negative examples, should not be taken.
In this way, they have been able to improve an artificial intelligence algorithm that can be used in areas such as finding drugs and optimizing logistics management.
Christian BlumThe scientist responsible for this work by the IIIA-CSIC explained: “The type of learning used in ACO is limited to learning from positive examples. However, please ask Negative example“It seems to play an important role in the self-organizing biological system. Pharaoh ants use negative pheromone to show signs of no entry, thereby marking unhelpful eating paths.”
At work, the doctoral students are also signatories Teddy Nurcahyadi, Designed the first general mechanism that beneficially incorporates negative learning into ACO technology.Submitted in October ANTS 2020 Conference The study is one of Barcelona’s main research objects, and the study was rated as the best article due to its great potential for innovation.
Sum of positive and negative learning
The author has modified the ACO algorithm to merge learning based on negative examples: “This type of learning supplements active learning, which is still the most important, but in our article, we prove that the two together form a kind of A better algorithm.”
Blum explained: “Our algorithm is iterative, that is, the same instructions will be executed over and over again. “In each iteration, there seems to be a certain number of ants, and each ant will solve the problem. The problem generates possible effective solutions. “
If ants are essentially guided by the pheromone they find in each part of the road in a probabilistic way, then these pheromones in the algorithm will be equal to the values in the components of the possible solution. Like pheromone, these values will be enhanced positively or negatively depending on whether they appear in a good solution.
This type of algorithm can be applied to many optimization problems. The researchers said that in such a situation, there are many possible solutions, that is, to find the best solution, or at least to find a “good enough” solution, such as: Looking for a molecular combination of new drugs or logistics. “Without the right optimization tools, it is impossible to conduct research in many areas,” Bloom concludes.
Teddy Nurcahyadi, Christian Blum. “A new method of using negative learning in ant colony optimization”.Proceedings ANTS 2020: Group Intelligence
This research is carried out in the project of the national R+D+i program CI maintenance: Advanced computational intelligence to achieve the goal of sustainable development.