Karush Suri

Philip Fradkin, Puria Azadi, Karush Suri et al.

Predicting molecular impact on cellular function is a core challenge in therapeutic design. Phenomic experiments, designed to capture cellular morphology, utilize microscopy based techniques and demonstrate a high throughput solution for uncovering molecular impact on the cell. In this work, we learn a joint latent space between molecular structures and microscopy phenomic experiments, aligning paired samples with contrastive learning. Specifically, we study the problem ofContrastive PhenoMolecular Retrieval, which consists of zero-shot molecular structure identification conditioned on phenomic experiments. We assess challenges in multi-modal learning of phenomics and molecular modalities such as experimental batch effect, inactive molecule perturbations, and encoding perturbation concentration. We demonstrate improved multi-modal learner retrieval through (1) a uni-modal pre-trained phenomics model, (2) a novel inter sample similarity aware loss, and (3) models conditioned on a representation of molecular concentration. Following this recipe, we propose MolPhenix, a molecular phenomics model. MolPhenix leverages a pre-trained phenomics model to demonstrate significant performance gains across perturbation concentrations, molecular scaffolds, and activity thresholds. In particular, we demonstrate an 8.1x improvement in zero shot molecular retrieval of active molecules over the previous state-of-the-art, reaching 77.33% in top-1% accuracy. These results open the door for machine learning to be applied in virtual phenomics screening, which can significantly benefit drug discovery applications.

Maciej Sypetkowski, Frederik Wenkel, Farimah Poursafaei et al.

Scaling deep learning models has been at the heart of recent revolutions in language modelling and image generation. Practitioners have observed a strong relationship between model size, dataset size, and performance. However, structure-based architectures such as Graph Neural Networks (GNNs) are yet to show the benefits of scale mainly due to the lower efficiency of sparse operations, large data requirements, and lack of clarity about the effectiveness of various architectures. We address this drawback of GNNs by studying their scaling behavior. Specifically, we analyze message-passing networks, graph Transformers, and hybrid architectures on the largest public collection of 2D molecular graphs. For the first time, we observe that GNNs benefit tremendously from the increasing scale of depth, width, number of molecules, number of labels, and the diversity in the pretraining datasets. We further demonstrate strong finetuning scaling behavior on 38 highly competitive downstream tasks, outclassing previous large models. This gives rise to MolGPS, a new graph foundation model that allows to navigate the chemical space, outperforming the previous state-of-the-arts on 26 out the 38 downstream tasks. We hope that our work paves the way for an era where foundational GNNs drive pharmaceutical drug discovery.

Ihab Bendidi, Yassir El Mesbahi, Alisandra K. Denton et al.

Understanding cellular responses to stimuli is crucial for biological discovery and drug development. Transcriptomics provides interpretable, gene-level insights, while microscopy imaging offers rich predictive features but is harder to interpret. Weakly paired datasets, where samples share biological states, enable multimodal learning but are scarce, limiting their utility for training and multimodal inference. We propose a framework to enhance transcriptomics by distilling knowledge from microscopy images. Using weakly paired data, our method aligns and binds modalities, enriching gene expression representations with morphological information. To address data scarcity, we introduce (1) Semi-Clipped, an adaptation of CLIP for cross-modal distillation using pretrained foundation models, achieving state-of-the-art results, and (2) PEA (Perturbation Embedding Augmentation), a novel augmentation technique that enhances transcriptomics data while preserving inherent biological information. These strategies improve the predictive power and retain the interpretability of transcriptomics, enabling rich unimodal representations for complex biological tasks.

Karush Suri, Xiao Qi Shi, Konstantinos Plataniotis et al.

Advances in Reinforcement Learning (RL) span a wide variety of applications which motivate development in this area. While application tasks serve as suitable benchmarks for real world problems, RL is seldomly used in practical scenarios consisting of abrupt dynamics. This allows one to rethink the problem setup in light of practical challenges. We present Trade Execution using Reinforcement Learning (TradeR) which aims to address two such practical challenges of catastrophy and surprise minimization by formulating trading as a real-world hierarchical RL problem. Through this lens, TradeR makes use of hierarchical RL to execute trade bids on high frequency real market experiences comprising of abrupt price variations during the 2019 fiscal year COVID19 stock market crash. The framework utilizes an energy-based scheme in conjunction with surprise value function for estimating and minimizing surprise. In a large-scale study of 35 stock symbols from the S&P500 index, TradeR demonstrates robustness to abrupt price changes and catastrophic losses while maintaining profitable outcomes. We hope that our work serves as a motivating example for application of RL to practical problems.

Karush Suri, Xiao Qi Shi, Konstantinos Plataniotis et al.

Multi-Agent Reinforcement Learning (MARL) has demonstrated significant success in training decentralised policies in a centralised manner by making use of value factorization methods. However, addressing surprise across spurious states and approximation bias remain open problems for multi-agent settings. Towards this goal, we introduce the Energy-based MIXer (EMIX), an algorithm which minimizes surprise utilizing the energy across agents. Our contributions are threefold; (1) EMIX introduces a novel surprise minimization technique across multiple agents in the case of multi-agent partially-observable settings. (2) EMIX highlights a practical use of energy functions in MARL with theoretical guarantees and experiment validations of the energy operator. Lastly, (3) EMIX extends Maxmin Q-learning for addressing overestimation bias across agents in MARL. In a study of challenging StarCraft II micromanagement scenarios, EMIX demonstrates consistent stable performance for multiagent surprise minimization. Moreover, our ablation study highlights the necessity of the energy-based scheme and the need for elimination of overestimation bias in MARL. Our implementation of EMIX can be found at karush17.github.io/emix-web/.

Karush Suri, Xiao Qi Shi, Konstantinos N. Plataniotis et al.

Advances in Reinforcement Learning (RL) have demonstrated data efficiency and optimal control over large state spaces at the cost of scalable performance. Genetic methods, on the other hand, provide scalability but depict hyperparameter sensitivity towards evolutionary operations. However, a combination of the two methods has recently demonstrated success in scaling RL agents to high-dimensional action spaces. Parallel to recent developments, we present the Evolution-based Soft Actor-Critic (ESAC), a scalable RL algorithm. We abstract exploration from exploitation by combining Evolution Strategies (ES) with Soft Actor-Critic (SAC). Through this lens, we enable dominant skill transfer between offsprings by making use of soft winner selections and genetic crossovers in hindsight and simultaneously improve hyperparameter sensitivity in evolutions using the novel Automatic Mutation Tuning (AMT). AMT gradually replaces the entropy framework of SAC allowing the population to succeed at the task while acting as randomly as possible, without making use of backpropagation updates. In a study of challenging locomotion tasks consisting of high-dimensional action spaces and sparse rewards, ESAC demonstrates improved performance and sample efficiency in comparison to the Maximum Entropy framework. Additionally, ESAC presents efficacious use of hardware resources and algorithm overhead. A complete implementation of ESAC can be found at karush17.github.io/esac-web/.

Karush Suri, Rinki Gupta

Sign Language is used by the deaf community all over world. The work presented here proposes a novel one-dimensional deep capsule network (CapsNet) architecture for continuous Indian Sign Language recognition by means of signals obtained from a custom designed wearable IMU system. The performance of the proposed CapsNet architecture is assessed by altering dynamic routing between capsule layers. The proposed CapsNet yields improved accuracy values of 94% for 3 routings and 92.50% for 5 routings in comparison with the convolutional neural network (CNN) that yields an accuracy of 87.99%. Improved learning of the proposed architecture is also validated by spatial activations depicting excited units at the predictive layer. Finally, a novel non-cooperative pick-and-predict competition is designed between CapsNet and CNN. Higher value of Nash equilibrium for CapsNet as compared to CNN indicates the suitability of the proposed approach.

Karush Suri, Rinki Gupta

Recent advancements in diagnostic learning and development of gesture-based human machine interfaces have driven surface electromyography (sEMG) towards significant importance. Analysis of hand gestures requires an accurate assessment of sEMG signals. The proposed work presents a novel sequential master-slave architecture consisting of deep neural networks (DNNs) for classification of signs from the Indian sign language using signals recorded from multiple sEMG channels. The performance of the master-slave network is augmented by leveraging additional synthetic feature data generated by long short term memory networks. Performance of the proposed network is compared to that of a conventional DNN prior to and after the addition of synthetic data. Up to 14% improvement is observed in the conventional DNN and up to 9% improvement in master-slave network on addition of synthetic data with an average accuracy value of 93.5% asserting the suitability of the proposed approach.

Rinki Gupta, Karush Suri

Surface electromyography (sEMG) has gained significant importance during recent advancements in consumer electronics for healthcare systems, gesture analysis and recognition and sign language communication. For such a system, it is imperative to determine the regions of activity in a continuously recorded sEMG signal. The proposed work provides a novel activity detection approach based on Hidden Markov Models (HMM) using sEMG signals recorded when various hand gestures are performed. Detection procedure is designed based on a probabilistic outlook by making use of mathematical models. The requirement of a threshold for activity detection is obviated making it subject and activity independent. Correctness of the predicted outputs is asserted by classifying the signal segments around the detected transition regions as activity or rest. Classified outputs are compared with the transition regions in a stimulus given to the subject to perform the activity. The activity onsets are detected with an average of 96.25% accuracy whereas the activity termination regions with an average of 87.5% accuracy with the considered set of six activities and four subjects.

Karush Suri, Rinki Gupta

Surface electromyography (sEMG) is becoming exceeding useful in applications involving analysis of human motion such as in human-machine interface, assistive technology, healthcare and prosthetic development. The proposed work presents a novel dual stage classification approach for classification of grasping gestures from sEMG signals. A statistical assessment of these activities is presented to determine the similar characteristics between the considered activities. Similar activities are grouped together. In the first stage of classification, an activity is identified as belonging to a group, which is then further classified as one of the activities within the group in the second stage of classification. The performance of the proposed approach is compared to the conventional single stage classification approach in terms of classification accuracies. The classification accuracies obtained using the proposed dual stage classification are significantly higher as compared to that for single stage classification.

Karush Suri, Rinki Gupta

IMUs are gaining significant importance in the field of hand gesture analysis, trajectory detection and kinematic functional study. An Inertial Measurement Unit (IMU) consists of tri-axial accelerometers and gyroscopes which can together be used for formation analysis. The paper presents a novel classification approach using a Deep Neural Network (DNN) for classifying hand gestures obtained from wearable IMU sensors. An optimization objective is set for the classifier in order to reduce correlation between the activities and fit the signal-set with best performance parameters. Training of the network is carried out by feed-forward computation of the input features followed by the back-propagation of errors. The predicted outputs are analyzed in the form of classification accuracies which are then compared to the conventional classification schemes of SVM and kNN. A 3-5% improvement in accuracies is observed in the case of DNN classification. Results are presented for the recorded accelerometer and gyroscope signals and the considered classification schemes.

Karush Suri, Rinki Gupta

Advancements in gesture recognition algorithms have led to a significant growth in sign language translation. By making use of efficient intelligent models, signs can be recognized with precision. The proposed work presents a novel one-dimensional Convolutional Neural Network (CNN) array architecture for recognition of signs from the Indian sign language using signals recorded from a custom designed wearable IMU device. The IMU device makes use of tri-axial accelerometer and gyroscope. The signals recorded using the IMU device are segregated on the basis of their context, such as whether they correspond to signing for a general sentence or an interrogative sentence. The array comprises of two individual CNNs, one classifying the general sentences and the other classifying the interrogative sentence. Performances of individual CNNs in the array architecture are compared to that of a conventional CNN classifying the unsegregated dataset. Peak classification accuracies of 94.20% for general sentences and 95.00% for interrogative sentences achieved with the proposed CNN array in comparison to 93.50% for conventional CNN assert the suitability of the proposed approach.