Diffusion models have demonstrated significant potential in achieving state-of-the-art performance across various text generation tasks. In this systematic study, we investigate their application to the table-to-text problem by adapting the diffusion model to the task and conducting an in-depth analysis. Our experiments cover multiple aspects of diffusion models training. We explore sampling strategy influence by inducing recent diffusion model accelerator DPM-Solver++ into our core model. We have tested different prediction aggregation methods, like ROVER and Minimum Bayes-Risk (MBR). Our studies cover the impact of the pre-training phase in diffusion models and the generation length constraints influence. We also have compared diffusion model generation with auto-regressive text-to-text models with different temperature settings for diversity evaluation. Our key observation is that diffusion models demonstrate the balance between quality and diversity while auto-regressive text-to-text models are not successful at handling both at the same time. Furthermore, we found out that to achieve the highest quality possible, it is preferable to use a regular sampler with the strictest length constraint to create multiple samples, and then use MBR to aggregate the predictions. However, if you are prepared to give up high level of diversity and to accelerate the process, you can also utilize a fast sampler DPM-Solver++. Our findings reveal that diffusion models achieve comparable results in the table-to-text domain, highlighting their viability in the table-to-text challenge as a promising research direction.

IMAGE: Researchers from the Moscow Institute of Physics and Technology, Ivannikov Institute for System Programming, and the Harvard Medical School-affiliated Schepens Eye Research Institute have developed a neural network capable of... view more Researchers from the Moscow Institute of Physics and Technology, Ivannikov Institute for System Programming, and the Harvard Medical School-affiliated Schepens Eye Research Institute have developed a neural network capable of recognizing retinal tissues during the process of their differentiation in a dish. Unlike humans, the algorithm achieves this without the need to modify cells, making the method suitable for growing retinal tissue for developing cell replacement therapies to treat blindness and conducting research into new drugs. The study was published in Frontiers in Cellular Neuroscience. In multicellular organisms, the cells making up different organs and tissues are not the same. They have distinct functions and properties, acquired in the course of development.

Researchers from Russian corporation Neurobotics and the Moscow Institute of Physics and Technology have found a way to visualize a person's brain activity as actual images mimicking what they observe in real time. This will enable new post-stroke rehabilitation devices controlled by brain signals. The team published its research as a preprint on bioRxiv and posted a video online, showing their "mind-reading" system at work. To develop devices controlled by the brain and methods for cognitive disorder treatment and post-stroke rehabilitation, neurobiologists need to understand how the brain encodes information. A key aspect of this is studying the brain activity of people perceiving visual information, for example, while watching a video.

The Otto Group Product Classification Challenge made Kaggle history as our most popular competition ever. Alexander Guschin finished in 2nd place ahead of 3,845 other data scientists. In this blog, Alexander shares his stacking centered approach and explains why you should never underestimate the nearest neighbours algorithm. I have some theoretical understanding of machine learning thanks to my base institute (Moscow Institute of Physics and Technology) and our professor Konstantin Vorontsov, one of the top Russian machine learning specialists. As for my acquaintance with practical problems, another great Russian data scientist who once was Top-1 on Kaggle, Alexander D'yakonov, used to teach a course on practical machine learning every autumn which gave me very good basis. Kagglers may know this course as PZAD.

A team of scientists from the Moscow Institute of Physics and Technology (MIPT) have created prototypes of "electronic synapses" based on ultra-thin films of hafnium oxide (HfO2). These prototypes could potentially be used in fundamentally new computing systems. The paper has been published in the journal Nanoscale Research Letters. The group of researchers from MIPT have made HfO2-based memristors measuring just 40x40 nm2. The nanostructures they built exhibit properties similar to biological synapses.