Dmitriy Gizlyk / Profil

We continue to study algorithms for extracting features from a point cloud. In this article, we will get acquainted with the mechanisms for increasing the efficiency of the PointNet method.

Direct point cloud analysis avoids unnecessary data growth and improves the performance of models in classification and segmentation tasks. Such approaches demonstrate high performance and robustness to perturbations in the original data.

We continue studying the Hierarchical Vector Transformer method. In this article, we will complete the construction of the model. We will also train and test it on real historical data.

We invite you to get acquainted with the Hierarchical Vector Transformer (HiVT) method, which was developed for fast and accurate forecasting of multimodal time series.

Understanding agent behavior is important in many different areas, but most methods focus on just one of the tasks (understanding, noise removal, or prediction), which reduces their effectiveness in real-world scenarios. In this article, we will get acquainted with a model that can adapt to solving various problems.

In this article, I would like to introduce you to an interesting trajectory prediction method developed to solve problems in the field of autonomous vehicle movements. The authors of the method combined the best elements of various architectural solutions.

A large number of the models we have reviewed so far are based on the Transformer architecture. However, they may be inefficient when dealing with long sequences. And in this article, we will get acquainted with an alternative direction of time series forecasting based on state space models.

Most modern multimodal time series forecasting methods use the independent channels approach. This ignores the natural dependence of different channels of the same time series. Smart use of two approaches (independent and mixed channels) is the key to improving the performance of the models.

We continue our acquaintance with the TEMPO method. In this article we will evaluate the actual effectiveness of the proposed approaches on real historical data.

We continue to study time series forecasting models. In this article, we get acquainted with a complex algorithm built on the use of a pre-trained language model.

Lightweight time series forecasting models achieve high performance using a minimum number of parameters. This, in turn, reduces the consumption of computing resources and speeds up decision-making. Despite being lightweight, such models achieve forecast quality comparable to more complex ones.

One of the directions for increasing the efficiency of the model training and convergence process is the improvement of optimization methods. Adam-mini is an adaptive optimization method designed to improve on the basic Adam algorithm.

In this article we will talk about using space-time transformations to effectively predict upcoming price movement. To improve the numerical prediction accuracy in STNN, a continuous attention mechanism is proposed that allows the model to better consider important aspects of the data.

We continue the discussion about the use of piecewise linear representation of time series, which was started in the previous article. Today we will see how to combine this method with other approaches to time series analysis to improve the price trend prediction quality.

This article is somewhat different from my earlier publications. In this article, we will talk about an alternative representation of time series. Piecewise linear representation of time series is a method of approximating a time series using linear functions over small intervals.

When exploring various model architecture designs, we often devote insufficient attention to the process of model training. In this article, I aim to address this gap.

Efficient extraction and integration of long-term dependencies and short-term features remain an important task in time series analysis. Their proper understanding and integration are necessary to create accurate and reliable predictive models.

Transformer architecture-based models demonstrate high efficiency, but their use is complicated by high resource costs both at the training stage and during operation. In this article, I propose to get acquainted with algorithms that allow to reduce memory usage of such models.

When working with time series, we always use the source data in their historical sequence. But is this the best option? There is an opinion that changing the sequence of the input data will improve the efficiency of the trained models. In this article I invite you to get acquainted with one of the methods for optimizing the input sequence.

In this article, we continue the implementation of the approaches of the ATFNet model, which adaptively combines the results of 2 blocks (frequency and time) within time series forecasting.