Artem Moskalev

Geometric Hyena Networks for Large-scale Equivariant Learning

Artem Moskalev, Mangal Prakash, Junjie Xu, Tianyu Cui, Rui Liao, Tommaso Mansi

ICML, Spotlight, 2025

Existing equivariant models are either computationally expensive or lose global structure. We introduce Geometric Hyena, the first equivariant long-convolutional model, combining sub-quadratic complexity with rotational and translational equivariance. On RNA property prediction and protein dynamics, it outperforms prior models while using far less memory and compute, processing 30k tokens 20× faster and enabling 72× longer context than equivariant transformers.

InfoSEM: A Deep Generative Model with Informative Priors for Gene Regulatory Network Inference

Tianyu Cui, Song-Jun Xu, Artem Moskalev, Shuwei Li, Tommaso Mansi, Mangal Prakash, Rui Liao

ICML, 2025

We present InfoSEM, an unsupervised model for GRN inference that uses pretrained gene embeddings and known interactions as priors. Unlike supervised methods that often exploit dataset-specific biases, InfoSEM captures true biological signals and achieves state-of-the-art performance on a biologically grounded benchmarks.

HARMONY: A Multi-Representation Framework for RNA Property Prediction

Junjie Xu, Artem Moskalev, Tommaso Mansi, Mangal Prakash, Rui Liao

ICLR: AI for Nucleic Acids, Oral, 2025

We introduce HARMONY, a neural network that dynamically integrates 1D, 2D, and 3D representations, and seamlessly adapts to diverse real-world scenarios. Our experiments demonstrate that HARMONY consistently outperforms existing baselines across multiple RNA property prediction tasks on established benchmarks, offering a robust and generalizable approach to RNA modeling.

HELM: Hierarchical Encoding for mRNA Language Modeling

Mehdi Yazdani-Jahromi, Mangal Prakash, Tommaso Mansi, Artem Moskalev*, Rui Liao*

ICLR, 2025

We introduce Hierarchical Encoding for mRNA Language Modeling (HELM), a novel pre-training strategy that incorporates codon-level hierarchical structure into language model training. HELM modulates the loss function based on codon synonymity, aligning the model's learning process with the biological reality of mRNA sequences.

Beyond Sequence: Impact of Geometric Context for RNA Property Prediction

Junjie Xu*, Artem Moskalev*, Tommaso Mansi, Mangal Prakash, Rui Liao

ICLR, 2025

We present the first systematic study of incorporating geometric context—beyond 1D sequences—into RNA property prediction. We reveal that geometry-aware models are more accurate while requiring less training data. At the same time, plain sequence-based models are the most robust to sequencing noise.

SE(3)-Hyena Operator for Scalable Equivariant Learning (Best Paper Award!!!)

Artem Moskalev, Mangal Prakash, Rui Liao, Tommaso Mansi

ICML: Geometry-grounded Representation Learning and Generative Modeling, 2024

We introduce SE(3)-Hyena operator, a translation and rotation equivariant long-convolutional method to process global geometric context at scale with sub-quadratic complexity. Significantly more compute and memory efficient than transformers.

On genuine invariance learning without weight-tying

Artem Moskalev, Anna Sepliarskaia, Erik J. Bekkers, Arnold Smeulders

ICML: Topology, Algebra, and Geometry in Machine Learning, 2023

We study properties and limitations of invariance learned by neural networks from the data compared to the invariance achieved through equivariant weight-tying. We next address the problem of aligning data-driven invariance learning to the genuine invariance of weight-tying models.

LieGG: Studying Learned Lie Group Generators

Artem Moskalev, Anna Sepliarskaia, Ivan Sosnovik, Arnold Smeulders

NeurIPS, Spotlight, 2022

We present LieGG, a method to extract symmetries learned by neural networks and to evaluate the degree to which a network is invariant to these symmetries. With LieGG, one can explicitly retrieve learned invariances in a form of the generators of corresponding Lie-groups without any prior knowledge of the symmetries in the data.

Contrasting quadratic assignments for set-based representation learning

Artem Moskalev, Ivan Sosnovik, Volker Fischer, Arnold Smeulders

ECCV, 2022

We go beyond contrasting individual pairs of objects by focusing on contrasting objects as sets. We use combinatorial quadratic assignment theory and derive set-contrastive objective as a regularizer for contrastive learning methods.

DISCO: accurate Discrete Scale Convolution (Best Paper Award!!!)

Ivan Sosnovik, Artem Moskalev, Arnold Smeulders

BMVC, Oral, 2021

We develop a better class of discrete scale equivariant CNNs, which are more accurate and faster than all previous methods. As a result of accurate scale analysis, they allow for a biased scene geometry estimation almost for free.

Relational Prior for Multi-Object Tracking

Artem Moskalev, Ivan Sosnovik, Arnold Smeulders

ICCV: VIPriors, Oral, 2021

Tracking multiple objects individually differs from tracking groups of related objects. We propose a plug-in Relation Encoding Module which encodes relations between tracked objects to improve multi-object tracking.

Scale Equivariance Improves Siamese Tracking

Ivan Sosnovik*, Artem Moskalev*, Arnold Smeulders

WACV, 2021

In this paper, we develop the theory for scale-equivariant Siamese trackers. We also provide a simple recipe for how to make a wide range of existing trackers scale-equivariant to capture the natural variations of the target a priori.