AI
Machine learning methods to protect banks from the risks of complex investment products
IMAGE:
PERFORMANCES OF THE PROPOSED METHOD COMPARED TO OPTIMAL STRATEGY
view moreCREDIT: CANNELLI, L., NUTI, G., SALA, M., AND SZEHR, O.
Artificial Intelligence (AI) is frequently touted as a silver bullet to solve complex modeling problems. Among its many applications, it has been investigated as a tool to manage risks of complex investment products—so-called derivative contracts—in the investment banking area. Despite the multiple positive reports in this area, concerns have been raised about their practical applicability.
In a new study published in The Journal of Finance and Data Science, a team of researchers from Switzerland and the US explored whether reinforcement learning RL agents can be trained to hedge derivative contracts.
"It should come as no surprise that if you train an AI on simulated market data, it will work well on markets that are reflective of the simulation, and the data consumption of many AI systems is outrageous," explains Loris Cannelli, first author of the study and a researcher at IDSIA in Switzerland.
To overcome the lack of training data, researchers tend to assume an accurate market simulator to train their AI agents. However, setting up such a simulator leads to a classical financial engineering problem: choosing a model to simulate from and its calibration, and making the AI-based approach much like the standard Monte Carlo methods in use for decades.
“Such an AI can also be hardly considered model-free: this would apply only if enough market data was available for training, and this is rarely the case in realistic derivative markets,” says Cannelli.
The study, a collaboration between IDSIA and investment bank of UBS, was based on so-called Deep Contextual Bandits, which are well-known in RL for their data-efficiency and robustness. Motivated by operational realities of real-world investment firms, it incorporates end of day reporting requirements and is characterized by a significantly lower training data requirement compared to conventional models, and adaptability to the changing markets.
"In practice, it's the availability of data and operational realities, such as requirements to report end-of-day risk figures, that are the main drivers that dictate the real work at the bank, instead of ideal agent training," clarifies senior author Oleg Szehr, whom, prior to his appointment at IDSIA, was a staff member at several investment banks. “One of the strengths of the newly developed model is that it conceptually resembles business operations at an investment firm and thus is applicable from a practical perspective.”
Although the new method is simple, rigorous assessment of model performance demonstrated that the new method outperforms benchmark systems in terms of efficiency, adaptability and accuracy under realistic conditions. “As often the case in real life, less is more—the same applies to risk management too,” concludes Cannelli.
###
Contact the author: Oleg Szehr, Dalle Molle Institute for Artificial Intelligence (IDSIA) – SUPSI/USI, oleg.szehr@supsi.ch
The publisher KeAi was established by Elsevier and China Science Publishing & Media Ltd to unfold quality research globally. In 2013, our focus shifted to open access publishing. We now proudly publish more than 100 world-class, open access, English language journals, spanning all scientific disciplines. Many of these are titles we publish in partnership with prestigious societies and academic institutions, such as the National Natural Science Foundation of China (NSFC).
JOURNAL
The Journal of Finance and Data Science
METHOD OF RESEARCH
Computational simulation/modeling
SUBJECT OF RESEARCH
Not applicable
ARTICLE TITLE
Hedging using reinforcement learning: Contextual k-armed bandit versus Q-learning
Artificial intelligence as therapeutic support
The face is a mirror for a person’s emotional state. The interpretation of facial expressions as part of psychotherapy or psychotherapeutic research, for example, is a very effective way of characterizing how a person is feeling in that particular moment. Back in the 1970s, psychologist Paul Ekmann developed a standardized coding system to assign basic emotions such as happiness, disgust or sadness to a facial expression in an image or video sequence.
“Ekman’s system is very widespread, and represents a standard in psychological emotion research,” says Dr. Martin Steppan, psychologist at the Faculty of Psychology at the University of Basel.
But the process of analyzing and interpreting recorded facial expressions as part of research projects or psychotherapy is extremely time-consuming, which is why psychiatry specialists often use less reliable, indirect methods such as skin conductance measurements, which can also be a measure of emotional arousal.
“We wanted to find out whether AI systems can reliably determine the emotional states of patients in video recordings,” says Martin Steppan, who developed the study together with emeritus Professor Klaus Schmeck, Dr. Ronan Zimmermann and Dr. Lukas Fürer from the UPK. The researchers published their findings in the journal Psychopathology.
No facial expression can escape AI
The researchers used freely available artificial neural networks that were trained in the detection of six basic emotions (happiness, surprise, anger, disgust, sadness and fear) using over 30,000 facial photos. This AI system then analyzed video data from therapy sessions with a total of 23 patients with borderline personality pathology at the Center for Scientific Computing at the University of Basel. The high-performance computer had to process over 950 hours of video recordings for this study.
The results were astonishing: statistical comparisons between the analysis of three trained therapists and the AI system showed a remarkable level of agreement. The AI system assessed the facial expressions as reliably as a human but was also able to detect even the most fleeting of emotions within the millisecond range, such as a brief smile or expression of disgust.
The results were astonishing: statistical comparisons between the analysis of three trained therapists and the AI system showed a remarkable level of agreement. The AI system assessed the facial expressions as reliably as a human but was also able to detect even the most fleeting of emotions within the millisecond range, such as a brief smile or expression of disgust.
These types of micro expressions have the potential to be missed by therapists or may only be perceived subconsciously. The AI system is therefore able to measure fleeting emotions with an increased level of sensitivity compared to trained therapists.
Interpersonal communication is still key
The AI analysis also uncovered something rather unexpected. Patients who demonstrated emotional involvement and smiled at the start of a therapy session went on to cancel their psychotherapy less often than people who seemed emotionally uninvolved with their therapist. This “social” smiling could therefore be a good predictor of therapy success in a person with symptoms of borderline personality pathology.
“We were really surprised to find that relatively simple AI systems can allocate facial expressions to their emotional states so reliably,” says Martin Steppan.
AI could therefore become an important tool in therapy and research. AI systems could be used in the analysis of existing video recordings from research studies in order to detect emotionally relevant moments in a conversation more easily and more directly. This ability could also help support the supervision of psychotherapists.
“Nevertheless, therapeutic work is still primarily about human relationships, and remains a human domain,” says Steppan. “At least for the time being.”
JOURNAL
Psychopathology
ARTICLE TITLE
Machine Learning Facial Emotion Classifiers in Psychotherapy Research: A Proof-of-Concept Study
Organic chemistry research transformed: The convergence of automation and AI reshapes scientific exploration
IMAGE:
(A) APPRAISAL OF THE RESEARCH GROUP’S DIVERSE INPUTS IN AI APPLICATIONS FOR ORGANIC CHEMISTRY. VISUALIZATION THROUGH (B) RESEARCH GROUPS’ AND (C) INSTITUTE’S WORD CLOUD MAPS, ALONG WITH (D) GEOGRAPHICAL DISTRIBUTION.
view moreCREDIT: ©SCIENCE CHINA PRESS
Recently, National Science Open magazine published online a review article led by Professor Fanyang Mo (School of Materials Science and Engineering, Peking University) and Professor Yuntian Chen (Eastern Institute of Technology, Ningbo). The research team proposed a significant shift towards automation and artificial intelligence (AI) in organic chemistry over the past decade. Furthermore, they introduced an innovative concept: the development of a generative, self-evolving AI chemistry research assistant.
The landscape of research in organic chemistry has undergone profound changes. Data, computing power, and sophisticated algorithms constitute the foundational pillars of AI-driven scientific research. In recent years, the rapid advancements in computing technology, coupled with the iterative enhancement of algorithms, have initiated a series of paradigm shifts in the scientific domain. This has led to a complete overhaul of conventional research methodologies. Organic chemistry, inherently predisposed to creating new substances, is uniquely positioned to thrive in this era of intelligent innovation. Scientists globally are now converging in their efforts to explore and harness the capabilities of artificial intelligence in chemistry, thus igniting the 'artificial intelligence chemistry' movement.
The academic realm is currently at the forefront of a research renaissance in this domain. The future holds great promise for the application of knowledge embedding and knowledge discovery techniques in scientific machine learning. This innovative approach is designed to narrow the gap between existing predictive models and automated experimental platforms, thereby facilitating the development of self-evolving AI chemical research assistants. In the field of organic chemistry, the concept of knowledge discovery through scientific machine learning is unlocking new possibilities. At the heart of this discipline is the understanding of reaction mechanisms, which often involve complex networks of intermediates, transition states, and concurrent reactions. Traditional approaches to deciphering these mechanisms have depended on kinetic studies and isotope labeling. However, merging symbolic mathematics with AI is poised to cast new light on these intricate pathways, potentially transforming both the understanding and teaching of organic chemical reactions.
Furthermore, the aspect of knowledge embedding holds significant importance from an organic chemist's perspective. Organic chemistry is replete with heuristic rules, ranging from Markovnikov's rules for electrophilic addition to Baldwin's rules for ring closures. Embedding these established principles into AI models would ensure that their predictions are not solely data-driven but also resonate with the intuitive understanding of chemists. This integration would yield insights that are both deeper and more aligned with the nuanced perspectives of organic chemistry.
Pipeline of generating self-evolving AI chemistry research assistant (IMAGE)
Data produced by advanced robotic systems employing high-throughput methods is processed using artificial intelligence models for decision-making. This dataset undergoes processes of knowledge discovery (green line) and embedding (blue line), enabling the enrichment of the AI's understanding. Subsequently, the AI model undergoes iterative updates, forming a continuous feedback loop that enhances its performance and decision-making capabilities
CREDIT
©Science China Press
See the article:
Transforming organic chemistry research paradigms: moving from manual efforts to the intersection of automation and artificial intelligence
https://doi.org/10.1360/nso/20230037
JOURNAL
National Science Open
No comments:
Post a Comment