Understanding cellular responses to chemical interventions is critical to the discovery of effective therapeutics. Because individual biological techniques often measure only one axis of cellular response at a time, high-quality multimodal datasets are needed to unlock a holistic understanding of how cells respond to treatments and to advance computational methods that integrate modalities. However, many techniques destroy cells and thus preclude paired measurements, and attempts to match disparate unimodal datasets are often confounded by data being generated in incompatible experimental settings. Here we introduce scGeneScope, a multimodal single‑cell RNA sequencing (scRNA-seq) and Cell Painting microscopy image dataset conditionally paired by chemical treatment, designed to facilitate the development and benchmarking of unimodal, multimodal, and multiple profile machine learning methods for cellular profiling. 28 chemicals, each acting on distinct biological pathways or mechanisms of action (MoAs), were applied to U2-OS cells in two experimental data generation rounds, creating paired sets of replicates that were then profiled independently by scRNA‑seq or Cell Painting. Using scGeneScope, we derive a replicate- and experiment-split treatment identification benchmark simulating MoA discovery under realistic laboratory variability conditions and evaluate unimodal, multimodal, and multiprofile models ranging in complexity from linear approaches to recent foundation models. Multiprofile integration improved performance in both the unimodal and multimodal settings, with gains more consistent in the former. Evaluation of unimodal models for MoA identification demonstrated that recent scRNA-seq foundation models deployed zero-shot were consistently outperformed by classic fit-to-data methods, underscoring the need for careful, realistic benchmarking in machine learning for biology. We release the scGeneScope dataset and benchmarking code to support further research.

We introduce a comprehensive framework for modeling single cell transcriptomic responses to perturbations, aimed at standardizing benchmarking in this rapidly evolving field. Our approach includes a modular and user-friendly model development and evaluation platform, a collection of diverse perturbational datasets, and a set of metrics designed to fairly compare models and dissect their performance. Through extensive evaluation of both published and baseline models across diverse datasets, we highlight the limitations of widely used models, such as mode collapse. We also demonstrate the importance of rank metrics which complement traditional model fit measures, such as RMSE, for validating model effectiveness. Notably, our results show that while no single model architecture clearly outperforms others, simpler architectures are generally competitive and scale well with larger datasets. Overall, this benchmarking exercise sets new standards for model evaluation, supports robust model development, and furthers the use of these models to simulate genetic and chemical screens for therapeutic discovery.

Class imbalance is a well-recognized challenge in GNN-based fraud detection. Traditional methods like re-sampling and re-weighting address this issue by balancing class distribution. However, node class balancing with simple re-sampling or re-weighting may greatly distort the data distributions and eventually lead to the ineffective performance of GNNs. In this paper, we propose a novel approach named Graph Generative Node Augmentation (GGA), which improves GNN-based fraud detection models by augmenting synthetic nodes of the minority class. GGA utilizes the GAN framework to synthesize node features and related edges of fake fraudulent nodes. To introduce greater variety in the generated nodes, we employ an MLP for feature generation. We also introduce an attention module to encode feature-level information before graph convolutional layers for edge generation. Our empirical results on two real-world fraud datasets demonstrate that GGA improves the performance of GNN-based fraud detection models by a large margin with much fewer nodes than traditional class balance methods, and outperforms recent graph augmentation methods with the same number of synthetic nodes.

Dynamic scene graph generation from a video is challenging due to the temporal dynamics of the scene and the inherent temporal fluctuations of predictions. We hypothesize that capturing long-term temporal dependencies is the key to effective generation of dynamic scene graphs. We propose to learn the long-term dependencies in a video by capturing the object-level consistency and inter-object relationship dynamics over object-level long-term tracklets using transformers. Experimental results demonstrate that our Dynamic Scene Graph Detection Transformer (DSG- DETR) outperforms state-of-the-art methods by a significant margin on the benchmark dataset Action Genome. Our ablation studies validate the effectiveness of each component of the proposed approach. The source code is available at https://github.com/Shengyu-Feng/DS G-DETR.

Being able to predict the performance of circuits without running expensive simulations is a desired capability that can catalyze automated design. In this article, we present a supervised pretraining approach to learn circuit representations that can be adapted to new circuit topologies or unseen prediction tasks. We hypothesize that if we train a neural network (NN) that can predict the output direct current (dc) voltages of a wide range of circuit instances it will be forced to learn generalizable knowledge about the role of each circuit element and how they interact with each other. The dataset for this supervised learning objective can be easily collected at scale since the required dc simulation to get ground truth labels is relatively cheap. This representation would then be helpful for few-shot generalization to unseen circuit metrics that require more time-consuming simulations for obtaining the ground-truth labels. To cope with the variable topological structure of different circuits we describe each circuit as a graph and use graph NNs (GNNs) to learn node embeddings. We show that pretraining GNNs on prediction of output node voltages can encourage learning representations that can be adapted to new unseen topologies or prediction of new circuit-level properties with up to 10x more sample efficiency compared to a randomly initialized model. We further show that we can improve the sample efficiency of prior SoTA model-based optimization methods by$2\times $(almost as good as using an oracle model) via fintuning pretrained GNNs as the feature extractor of the learned models.

Connection management is an important problem for any wireless network to ensure smooth and well-balanced operation throughout. Traditional methods for connection management (specifically user-cell association) consider sub-optimal and greedy solutions such as connection of each user to a cell with maximum receive power. However, network performance can be improved by leveraging machine learning (ML) and artificial intelligence (AI) based solutions. The next generation software defined 5G networks defined by the Open Radio Access Network (O-RAN) alliance facilitates the inclusion of ML/AI based solutions for various network problems. In this paper, we consider intelligent connection management based on the O-RAN network architecture to optimize user association and load balancing in the network. We formulate connection management as a combinatorial graph optimization problem. We propose a deep reinforcement learning (DRL) solution that uses the underlying graph to learn the weights of the graph neural networks (GNN) for optimal user-cell association. We consider three candidate objective functions: sum user throughput, cell coverage, and load balancing. Our results show up to 10% gain in throughput, 45-140% gain cell coverage, 20-45% gain in load balancing depending on network deployment configurations compared to baseline greedy techniques.

Recent improvements in the performance of state-of-the-art (SOTA) methods for Graph Representational Learning (GRL) have come at the cost of significant computational resource requirements for training, e.g., for calculating gradients via backprop over many data epochs. Meanwhile, Singular Value Decomposition (SVD) can find closed-form solutions to convex problems, using merely a handful of epochs. In this paper, we make GRL more computationally tractable for those with modest hardware. We design a framework that computes SVD of *implicitly* defined matrices, and apply this framework to several GRL tasks. For each task, we derive first-order approximation of a SOTA model, where we design (expensive-to-store) matrix $\mathbf{M}$ and train the model, in closed-form, via SVD of $\mathbf{M}$, without calculating entries of $\mathbf{M}$. By converging to a unique point in one step, and without calculating gradients, our models show competitive empirical test performance over various graphs such as article citation and biological interaction networks. More importantly, SVD can initialize a deeper model, that is architected to be non-linear almost everywhere, though behaves linearly when its parameters reside on a hyperplane, onto which SVD initializes. The deeper model can then be fine-tuned within only a few epochs. Overall, our algorithm trains hundreds of times faster than state-of-the-art methods, while competing on test empirical performance. We open-source our implementation at: https://github.com/samihaija/isvd

Academic citation graphs represent citation relationships between publications across the full range of academic fields. Top cited papers typically reveal future trends in their corresponding domains which is of importance to both researchers and practitioners. Prior citation prediction methods often require initial citation trends to be established and do not take advantage of the recent advancements in graph neural networks (GNNs). We present GNN-based architecture that predicts the top set of papers at the time of publication. For experiments, we curate a set of academic citation graphs for a variety of conferences and show that the proposed model outperforms other classic machine learning models in terms of the F1-score.

Deep learning has been undergoing rapid growth in recent years thanks to its state-of-the-art performance across a wide range of real-world applications. Traditionally neural networks were trained in IEEE-754 binary64 or binary32 format, a common practice in general scientific computing. However, the unique computational requirements of deep neural network training workloads allow for much more efficient and inexpensive alternatives, unleashing a new wave of numerical innovations powering specialized computing hardware. We previously presented Flexpoint, a blocked fixed-point data type combined with a novel predictive exponent management algorithm designed to support training of deep networks without modifications, aiming at a seamless replacement of the binary32 widely in practice today. We showed that Flexpoint with 16-bit mantissa and 5-bit shared exponent (flex16+S) achieved numerical parity to binary32 in training a number of convolutional neural networks. In the current paper we review the continuing trend of predictive numerics enhancing deep neural network training in specialized computing devices such as the Intel®N ervana ™ Neural Network Processor.

Deep neural networks are commonly developed and trained in 32-bit floating point format. Significant gains in performance and energy efficiency could be realized by training and inference in numerical formats optimized for deep learning. Despite advances in limited precision inference in recent years, training of neural networks in low bit-width remains a challenging problem. Here we present the Flexpoint data format, aiming at a complete replacement of 32-bit floating point format training and inference, designed to support modern deep network topologies without modifications. Flexpoint tensors have a shared exponent that is dynamically adjusted to minimize overflows and maximize available dynamic range. We validate Flexpoint by training AlexNet, a deep residual network and a generative adversarial network, using a simulator implemented with the \emph{neon} deep learning framework. We demonstrate that 16-bit Flexpoint closely matches 32-bit floating point in training all three models, without any need for tuning of model hyperparameters. Our results suggest Flexpoint as a promising numerical format for future hardware for training and inference.

Asynchronous impulsive noise and periodic impulsive noises limit communication performance in OFDM powerline communication systems. Conventional OFDM receivers that assume additive white Gaussian noise experience degradation in communication performance in impulsive noise. Alternate designs assume a statistical noise model and use the model parameters in mitigating impulsive noise. These receivers require training overhead for parameter estimation, and degrade due to model and parameter mismatch. To mitigate asynchronous impulsive noise, we exploit its sparsity in the time domain, and apply sparse Bayesian learning methods to estimate and subtract the noise impulses. We propose three iterative algorithms with different complexity vs. performance trade-offs: (1) we utilize the noise projection onto null and pilot tones; (2) we add the information in the date tones to perform joint noise estimation and symbol detection; (3) we use decision feedback from the decoder to further enhance the accuracy of noise estimation. These algorithms are also embedded in a time-domain block interleaving OFDM system to mitigate periodic impulsive noise. Compared to conventional OFDM receivers, the proposed methods achieve SNR gains of up to 9 dB in coded and 10 dB in uncoded systems in asynchronous impulsive noise, and up to 6 dB in coded systems in periodic impulsive noise.

A Smart Grid intelligently monitors and controls energy flows in an electric grid. Having up-to-date distributed readings of grid conditions helps utilities efficiently scale generation up or down to meet demand. Narrowband powerline communication (PLC) systems can provide these up-to-date readings from subscribers to the local utility over existing power lines. A key challenge in PLC systems is overcoming additive non-Gaussian noise. In this paper, we propose to use a cyclostationary model for the dominant component of additive non-Gaussian noise. The key contributions are (1) fitting measured data from outdoor narrowband PLC system field trials to a cyclostationary model, and (2) developing a cyclostationary noise generation model that fits measured data. We found that the period in the cyclostationary model matched half of the period of the main powerline frequency, which is consistent with previous work in indoor PLC additive noise modeling.

Additive asynchronous impulsive noise limits communication performance in certain OFDM systems, such as powerline communications, cellular LTE and 802.11n systems. Under additive impulsive noise, the fast Fourier transform (FFT) in the OFDM receiver introduces time-dependence in the subcarrier noise statistics. As a result, complexity of optimal detection becomes exponential in the number of subcarriers. Many previous approaches assume a statistical model of the impulsive noise and use parametric methods in the receiver to mitigate impulsive noise. Parametric methods degrade with increasing model mismatch, and require training and parameter estimation. In this paper, we apply sparse Bayesian learning techniques to estimate and mitigate impulsive noise in OFDM systems without the need for training. We propose two non-parametric iterative algorithms: (1) estimate impulsive noise by its projection onto null and pilot tones so that the OFDM symbol is recovered by subtracting out the impulsive noise estimate; and (2) jointly estimate the OFDM symbol and impulsive noise utilizing information on all tones. In our simulations, the estimators achieve 5dB and 10dB SNR gains in communication performance respectively, as compared to conventional OFDM receivers.

Powerline distribution networks are increasingly being employed to support smart grid communication infrastructure and in-home LAN connectivity. However, their primary function of power distribution results in a hostile environment for communication systems. In particular, asynchronous impulsive noise, with levels as high as 50dB above thermal noise, causes significant degradation in communication performance. Much of the prior work uses limited empirical measurements to propose a statistical model for instantaneous statistics of asynchronous noise. In this paper, we (i) derive a canonical statistical-physical model of the instantaneous statistics of asynchronous noise based on the physical properties of the PLC network, and (ii) validate the distribution using simulated and measured PLC noise data. The results of this paper can be used to analyze, simulate, and mitigate the effect of the asynchronous noise on PLC systems.

An IEEE 802.11b/g/n wireless local area network (WLAN) experiences significant radio frequency interference (RFI) from microwave ovens, cordless phones, Bluetooth, and other WLANs operating in the 2.4 GHz band. In the 2.4 GHz band, operating microwave ovens emit interference that can either prevent an access point from transmitting or cause dramatic increase in bit errors in the receiver. This paper reduces disruption in real-time transmission and reception of high-definition streaming video by mitigating oven-generated RFI. The proposed contributions of this paper include (1) a statistical-physical model of oven-generated RFI in WLAN channels that is dependent on distance and frequency, (2) an increase in the information rate bound of up to 5 bits/s/Hz due to the more realistic statistical-physical model of oven-generated RFI, and (3) a proposed transmission strategy to achieve the higher information rate bound.

In laptop and desktop computers, clocks and busses generate significant radio frequency interference (RFI) for the embedded wireless data transceivers. RFI is impulsive in nature. When detecting a signal in additive impulsive noise, Spaulding and Middleton showed a potential improvement in detection of 25 dB at a bit error rate of 10-5when using a Bayesian detector instead of a standard correlation receiver. In this paper, we model impulsive noise using Middleton class A and symmetric alpha stable (SaS) models. The contributions of this paper are to evaluate (1) the performance vs. complexity of parameter estimation algorithms, (2) the closeness of fit of parameter estimation algorithms to measured RFI data from the computer platform, (3) the communication performance vs. computational complexity tradeoffs for the correlation receiver, Wiener filter, and Bayesian detector, and (4) the performance of myriad filtering in combating RFI interference modeled as SaS interference.

A new method for multitask learning in a Bayesian network context is presented for multiorganism gene network estimation. When the input datasets are sparse, as is the case in microarray gene expression data, it becomes difficult to separate random correlations from actual edges in the true underlying Bayesian network. Multitask learning takes advantage of the similarity between related tasks, in order to construct a more accurate model of the underlying relationships represented by the Bayesian networks. The proposed method is tested on synthetic data to illustrate its validity. Then it is iteratively applied on real gene expression data to learn the genetic regulatory networks of two organisms with homologous genes (human and yeast).