P2P

90 P E E R T O P E E R : I L T A ' S Q U A R T E R L Y M A G A Z I N E | W I N T E R 2 0 2 3 Measuring AI Efficiency and Accuracy The decision to bring AI onboard your corporation's tech stack may not be a cheap one, and as with any other important acquisition it needs to be analyzed for risk and potential return. With AI, some of the metrics that decision- makers consider include precision, recall, processing speed, and AI training cost. High precision indicates that the AI system produces accurate results with a low rate of false positives. High recall suggests that the AI system can identify and retrieve most relevant information from a dataset. Processing speed affects how quickly an AI system can process large amounts of data; whereas training cost indicates the amount of money or time the corporation must spend on calibrating their AI system before they can begin realizing its full benefits. An ideal AI system would be one that requires minimal human supervision and can quickly learn and adapt to new data sources. Good performance across all the aforementioned metrics is critical for businesses that want to get the most out of their AI systems. High precision and high recall ensure the AI system produces accurate and relevant results, leading to better decision-making and improved business outcomes. High speed is essential for businesses that need to process large amounts of data quickly, such as in cases involving real-time compliance monitoring or rapid regulatory investigations, and time is, frankly, always of the essence. Cognizant of these priorities, FRONTEO has dedicated substantial development resources to continuously improve and iterate upon proprietary technology to create machine learning algorithms that deliver high-precision and high-recall results, while keeping training costs to a minimum. This paper will cover algorithmic performance improvements and provide further details in the sections below. Language-Driven Complexities The underlying performance of machine learning algorithms that drive a lot of the language-based analysis may be uneven across different languages. This should not be surprising – after all, syntax, grammar, and sentence structure can be quite varied even among related languages, not to mention something as dissimilar as Chinese and English. In fact, interpreting CJK (Chinese, Japanese, and Korean) languages with AI can be particularly challenging due to their complex writing systems. These languages are characterized by lack of space separation, high character counts and flexibility in word ordering. As an example, while English builds its words and sentences from only 26 letters, Chinese has over 20,000 characters (although many of them might be infrequently used), and Japanese sees over 2,000 characters in daily use. Additionally, many of these characters will have multiple meanings based on their context and position in a sentence, adding further complexity to accurate meaning interpretation. Another added challenge is that CJK languages have a unique grammar structure that differs significantly from English and other European languages. Chinese has no tense and no plural form of words. Both Chinese and Japanese have different word orders compared to English, and less rigid word order in a sentence. Korean has a complex system of particles that determines the grammatical function of a word in a sentence. These differences in syntax and grammar make it challenging to adapt AI models developed for Western markets to Asian data. KIBIT AI – Answer to Handling CJK content FRONTEO recognizes that developing accurate and effective AI systems for CJK languages requires extensive linguistic and cultural knowledge and poses unique F E A T U R E S

• Cover