Amin Beheshti’s Post

Financial markets are complex environments. Their complexity stems from two main factors: (i) the interaction of a large number of agents pursuing heterogeneous goals at different time scales through the implementation of trading strategies designed to leverage asymmetric information; (ii) the emergence of self-organizing collective behaviors that do not result from the existence of any central controller and are therefore difficult to anticipate. Understanding how variables evolve at the finest-grain resolution (i.e., order submission level) to influence the collective behavior of the underlying system and how the system, in turn, impacts individual variables is crucial to making the trading practice profitable. In the paper titled "HLOB - Information Persistence and Structure in Limit Order Books" (coauthored with Prof. Silvia Bartolucci and Prof. Tomaso Aste): - We demonstrate that Limit Order Books are characterized by spatial structures that impose the modeling of higher-order and non-trivial dependency structures among volume (and price) levels. - We introduce 'HLOB', a novel, large-scale deep learning architecture that exploits the Homological Convolutional Neural Networks class to superimpose a dependency structure among Limit Order Book volume levels and model complex relationships among them. - We show that the exploitability of the informational content encoded in the spatial structures imposed by our deep learning architecture is limited in time, and the velocity of its degradation is highly dependent on the stocks’ microstructural properties. Link to the paper: https://lnkd.in/gQ8_MikH Link to the code: https://lnkd.in/d47B-eQE #finance #markets #financialmarkets #ai #deeplearning #physics #econophysics #forecasting #timeseries #limitorderbook #computerscience

🎤 Speaker Spotlight 💫💡 Sebastian Raubitzek, SBA Research Chaos, Complexity and Neural Network Time Series Predictions 🖥 Many of today's most successful approaches for predicting time series data use #machinelearning and/or #deeplearning approaches such as different #neuralnetwork architectures. These approaches strongly depend on the data available to train the employed algorithm, long-term data sets are rare and often sparsely sampled. They are often difficult to predict because of numerous influences that affect these data sets. Thus, these data sets have an inherent randomness to them. The problem of sparsely sampled data and the inherent randomness can be overcome by employing different interpolation techniques as well as ensemble predictions. 🔍 The presented research introduces you to two novel interpolation techniques. A Hurst-exponent-based fractal interpolation taking into account the fluctuative nature of stochastic time series data. As well as a stochastic interpolation method that looks at the reconstructed phase space properties of chaotic time series to produce an interpolation with a smooth phase space trajectory. Furthermore, you will learn about an ensemble technique that takes into account the complexity and/or reconstructed phase-space properties of the data under study. This is achieved by randomly parameterizing a multitude of long short term memory neural networks (#LSTM), having them produce an autoregressive prediction, and afterwards filtering this multitude of different predictions based on their signal complexity and/or reconstructed phase space properties. The results show that neural network time-series predictions can drastically be improved by employing the discussed interpolation techniques. They can effectively be filtered based on their inherent complexity and phase space properties to improve ensemble predictions. 🕑🔬 👨🎓 Sebastian Raubitzek studied theoretical physics with a focus on Higgs physics in Graz. He then expanded his expertise with a PhD in Computer Science/AI. Since January 2023, he has been working at SBA-Research in the Complexity and Resilience Research Group to conduct research around complexity, chaos and artificial intelligence. 🤖🔍 #SAINT2024 #AI #artificialintelligence #KI #künstlicheintelligenz #datascience

📃Scientific paper: Parameter Estimation for Intermediate-Mass Binary Black Holes through Gravitational Waves Observed by DECIGO Abstract: With the anticipated launch of space-based gravitational wave detectors, including LISA, TaiJi, TianQin, and DECIGO, expected around 2030, the detection of gravitational waves generated by intermediate-mass black hole binaries (IMBBHs) becomes a tangible prospect. However, due to the detector's reception of a substantial amount of non-Gaussian, non-stationary data, employing traditional Bayesian inference methods for parameter estimation would result in significant resource demands and limitations in the waveform template library. Therefore, in this paper, we simulated foreground noise induced by stellar-origin binary black holes (SOBBHs), which is non-Gaussian and non-stationary, and we explore the use of Gaussian process regression (GPR) and deep learning for parameter estimation of Intermediate Mass Binary Black Holes (IMBBHs) in the presence of such non-Gaussian, non-stationary background noise. By comparing these results from deep learning and GPR, we demonstrate that deep learning can offer improved precision in parameter estimation compared to traditional GPR. Furthermore, compared to GPR, deep learning can provide posterior distributions of the sample parameters faster. Continued on ES/IODE ➡️ https://etcse.fr/H3AXI ------- If you find this interesting, feel free to follow, comment and share. We need your help to enhance our visibility, so that our platform continues to serve you.

Another application of deep learning by UH AQF Choi's group, first authored by Ph.D. student Mahsa Payami, involves developing a straightforward deep learning emulator (digital twin) to replicate the results of the EPA CMAQ air quality model using similar inputs. In terms of computational efficiency, this emulator is 900 times faster than the original numerical model. This approach can potentially be applied to create emulators for other air quality, weather, or climate models. https://lnkd.in/gyumf5E3

A new research released last week offering a replacement of one of the foundational building blocks of today’s deep learning model which is MLP. Inspired by the "Kolmogorov-Arnold representation theorem" the research introduced Kolmogorov-Arnold Networks (KANs), simply the difference between MLPs and KANs is that MLPs place fixed activation functions on neurons while KANs place learnable activation functions on edges (weights). While MLPs represents a transformed version of the weighted sum of the inputs KANs represents a sum of transformed version of the inputs and instead of learning the weights KANs learn the transformation function which can vary from neuron to neuron in the same layer. Using this way the authors claimed that KANs can outperform MLPs using smaller architecture leading to more accuracy, less latency. In summary, KANs introduce flexibility by allowing learnable activation functions associated with edges, potentially leading to more efficient and accurate models. Researchers are exploring this promising approach to enhance deep learning architectures. 😊🚀 Paper 📑 : https://lnkd.in/dQkM8ueE Code 💻 : https://lnkd.in/dWAE8Ti6 Docs 📕 : https://lnkd.in/dajhNg_9 #DeepLearning #Kolmogorov_Arnold #KAN #MLP

The integration of Machine Learning and Generative AI (deep learning) capabilities has real potential to do improve the ability to do predictions with real-world complexities. This paper shows ways to maintain the innate structures of physics and augment them with broader learning techniques. This is the way to think about the next generation of applications of AI. https://lnkd.in/gptmfJPk

In the first Keynote of #HHAI2024 Fosca Giannotti talks about the future of AI and about creating systems that effectively collaborate with humans, combining strengths to tackle complex problems beyond the capability of either alone. https://lnkd.in/eGhDiij2

I am delighted to share that our latest research, leveraging deep learning to enhance the accuracy of satellite data, has been published in Environment International. Despite significant advancements in satellite technology, particularly with instruments in geostationary orbit, persistent biases continue to challenge the accuracy of satellite-derived data. Our study specifically addressed these issues by employing deep learning techniques to correct biases in the tropospheric column density of NO2 (TCDNO2) as measured by the GEMS satellite instrument over the period from 2021 to 2023. The deep learning-corrected GEMS TCDNO2 (GEMS-DL) demonstrates a substantial improvement in the accuracy of the original GEMS TCDNO2 measurements. Validation, conducted using Pandora stations, reveals that our GEMS-DL product increases the correlation coefficient (R) from 0.68 to 0.88, improves the Index of Agreement (IOA) from 0.79 to 0.93, decreases the Mean Absolute Bias (MAB) from 5.73321×1015 to 2.67659×1015 molecules/cm2, and reduces the MAB percentage (MABP) from 64% to 30%. This represents a significant reduction in bias, exceeding 50%. While the original GEMS data tends to overestimate TCDNO2 by 28%, our GEMS-DL product remarkably minimizes this error to an underestimation of just 1%. Link to the article: https://lnkd.in/dRxJ8fxp #satellites #deeplearning #GEMS #Pandora

🚲Building on Nobel-Winning AI Foundations: My Work on Neural Networks for Urban Mobility🚲 With neural networks making headlines again thanks to the recent Nobel Prize in Physics awarded to John Hopfield and Geoffrey Hinton, it's an exciting moment for everyone involved in AI and machine learning. Their foundational work has paved the way for many advancements, including applications like those in my 2022 research paper, "Predicting Shared-Bike Routes with Geographic Information System and LSTM Algorithm." This study focused on tackling the challenges of shared-bike systems using big data and machine learning, specifically addressing urban transportation issues with a two-layer LSTM model. Using over a million shared-bike tracks, we developed models that help optimize bike parking and distribution, promoting efficient and sustainable urban transportation. 🌱 Reflecting on this work today, I'm proud of its continued relevance as cities worldwide look for smarter solutions to reduce traffic congestion and promote green travel options. 🌍 For those interested in urban mobility and data-driven solutions, you can read the full paper here: https://lnkd.in/gPF9NGqa DOI: 10.2991/978-94-6463-005-3_57 I’m excited to connect with others who are passionate about sustainable city planning and data science. Let’s drive change together! 🙌 #NobelPrize #UrbanMobility #DataScience #LSTM #GeographicInformationSystem #SustainableTransportation #NeuralNetworks

🌊 Revolutionizing Streamflow Forecasting with Causal Learning! 🌊 A groundbreaking study from Arizona State University and Tulane University introduces Causal Streamflow Forecasting (CSF), a novel approach that leverages spatio-temporal causal relationships to enhance the accuracy of streamflow predictions. 🔍 Key Highlights: - Traditional hydrologic models often struggle with the complex, nonlinear dynamics of streamflow. CSF utilizes spatio-temporal graph neural networks (STGNNs) to effectively capture these dynamics. - By integrating a river flow graph as prior knowledge, CSF respects causal relationships between upstream and downstream stations, leading to improved prediction accuracy. - Tested in the real-world context of Texas's Brazos River Basin, CSF outperforms conventional models across short, medium, and long-range forecasting tasks. 📈 Performance Metrics: CSF achieved remarkable results: - Nash-Sutcliffe Efficiency (NSE): 0.86 - Kling-Gupta Efficiency (KGE): 0.87 - Volumetric Efficiency (VE): 0.88 This innovative model not only demonstrates superior forecasting capabilities but also highlights the potential for integrating advanced neural network techniques with domain-specific knowledge in hydrologic modeling. 🌍 As we face increasing challenges in water resource management due to climate change and urbanization, advancements like these are crucial for sustainable planning. 🔗 Dive deeper into this transformative research that could reshape how we approach water resource management! #AI #Algorithms #ArtificialIntelligence #CausalInference #DL #DS #DataScience #DeepLearning #Hydrology #ML #MachineLearning #ResearchInnovation #StreamflowForecasting #Tech #Technology #WaterManagement Source: https://lnkd.in/eTyiEyzc