The Glass Finds from Kerem Maharal (pp. 75–83)

The Coins from Kerem Maharal (pp. 85–86)

The Flint Assemblage from Kerem Maharal (pp. 87–92)

Two cases of dermatomyositis with anti-cN1A antibody positivity

Dynamics of transcriptional programs and chromatin accessibility in mouse spermatogonial cells from early postnatal to adult life

In mammals, spermatogonial cells (SPGs) are undifferentiated male germ cells in testis that are quiescent until birth and then self-renew and differentiate to produce spermatogenic cells and functional sperm from early postnatal life throughout adulthood. The transcriptome of SPGs is highly dynamic and timely regulated during postnatal development. We examined if such dynamics involves changes in chromatin organization by profiling the transcriptome and chromatin accessibility of SPGs from early postnatal stages to adulthood in mice using deep RNA-seq, ATAC-seq and computational deconvolution analyses. By integrating transcriptomic and epigenomic features, we show that SPGs undergo massive chromatin remodeling during postnatal development that partially correlates with distinct gene expression profiles and transcription factors (TF) motif enrichment. We identify genomic regions with significantly different chromatin accessibility in adult SPGs that are marked by histone modifications associated with enhancers and promoters. Some of the regions with increased accessibility correspond to transposable element subtypes enriched in multiple TFs motifs and close to differentially expressed genes. Our results underscore the dynamics of chromatin organization in developing germ cells and complement existing datasets on SPGs by providing maps of the regulatory genome at high resolution from the same cell populations at early postnatal, late postnatal and adult stages collected from single individuals.

Hybrid deep learning approach for brain tumor classification using EfficientNetB0 and novel quantum genetic algorithm

Noisy Annotations in Semantic Segmentation

Obtaining accurate labels for instance segmentation is particularlychallenging due to the complex nature of the task. Each image necessitatesmultiple annotations, encompassing not only the object class but also itsprecise spatial boundaries. These requirements elevate the likelihood of errorsand inconsistencies in both manual and automated annotation processes. Bysimulating different noise conditions, we provide a realistic scenario forassessing the robustness and generalization capabilities of instancesegmentation models in different segmentation tasks, introducing COCO-N andCityscapes-N. We also propose a benchmark for weakly annotation noise, dubbedCOCO-WAN, which utilizes foundation models and weak annotations to simulatesemi-automated annotation tools and their noisy labels. This study sheds lighton the quality of segmentation masks produced by various models and challengesthe efficacy of popular methods designed to address learning with label noise.

How to Build a Quantum Supercomputer: Scaling from Hundreds to Millions of Qubits

In the span of four decades, quantum computation has evolved from anintellectual curiosity to a potentially realizable technology. Today,small-scale demonstrations have become possible for quantum algorithmicprimitives on hundreds of physical qubits and proof-of-principleerror-correction on a single logical qubit. Nevertheless, despite significantprogress and excitement, the path toward a full-stack scalable technology islargely unknown. There are significant outstanding quantum hardware,fabrication, software architecture, and algorithmic challenges that are eitherunresolved or overlooked. These issues could seriously undermine the arrival ofutility-scale quantum computers for the foreseeable future. Here, we provide acomprehensive review of these scaling challenges. We show how the road toscaling could be paved by adopting existing semiconductor technology to buildmuch higher-quality qubits, employing system engineering approaches, andperforming distributed quantum computation within heterogeneoushigh-performance computing infrastructures. These opportunities for researchand development could unlock certain promising applications, in particular,efficient quantum simulation/learning of quantum data generated by natural orengineered quantum systems. To estimate the true cost of such promises, weprovide a detailed resource and sensitivity analysis for classically hardquantum chemistry calculations on surface-code error-corrected quantumcomputers given current, target, and desired hardware specifications based onsuperconducting qubits, accounting for a realistic distribution of errors.Furthermore, we argue that, to tackle industry-scale classical optimization andmachine learning problems in a cost-effective manner, heterogeneousquantum-probabilistic computing with custom-designed accelerators should beconsidered as a complementary path toward scalability.

The Streetscape Application Services Stack (SASS): Towards a Distributed Sensing Architecture for Urban Applications

As urban populations grow, cities are becoming more complex, driving thedeployment of interconnected sensing systems to realize the vision of smartcities. These systems aim to improve safety, mobility, and quality of lifethrough applications that integrate diverse sensors with real-timedecision-making. Streetscape applications-focusing on challenges likepedestrian safety and adaptive traffic management-depend on managingdistributed, heterogeneous sensor data, aligning information across time andspace, and enabling real-time processing. These tasks are inherently complexand often difficult to scale. The Streetscape Application Services Stack (SASS)addresses these challenges with three core services: multimodal datasynchronization, spatiotemporal data fusion, and distributed edge computing. Bystructuring these capabilities as clear, composable abstractions with clearsemantics, SASS allows developers to scale streetscape applications efficientlywhile minimizing the complexity of multimodal integration. We evaluated SASS in two real-world testbed environments: a controlledparking lot and an urban intersection in a major U.S. city. These testbedsallowed us to test SASS under diverse conditions, demonstrating its practicalapplicability. The Multimodal Data Synchronization service reduced temporalmisalignment errors by 88%, achieving synchronization accuracy within 50milliseconds. Spatiotemporal Data Fusion service improved detection accuracyfor pedestrians and vehicles by over 10%, leveraging multicamera integration.The Distributed Edge Computing service increased system throughput by more thanan order of magnitude. Together, these results show how SASS provides theabstractions and performance needed to support real-time, scalable urbanapplications, bridging the gap between sensing infrastructure and actionablestreetscape intelligence.

ETS: Efficient Tree Search for Inference-Time Scaling

Test-time compute scaling has emerged as a new axis along which to improvemodel accuracy, where additional computation is used at inference time to allowthe model to think longer for more challenging problems. One promising approachfor test-time compute scaling is search against a process reward model, where amodel generates multiple potential candidates at each step of the search, andthese partial trajectories are then scored by a separate reward model in orderto guide the search process. The diversity of trajectories in the tree searchprocess affects the accuracy of the search, since increasing diversity promotesmore exploration. However, this diversity comes at a cost, as divergenttrajectories have less KV sharing, which means they consume more memory andslow down the search process. Previous search methods either do not performsufficient exploration, or else explore diverse trajectories but have highlatency. We address this challenge by proposing Efficient Tree Search (ETS),which promotes KV sharing by pruning redundant trajectories while maintainingnecessary diverse trajectories. ETS incorporates a linear programming costmodel to promote KV cache sharing by penalizing the number of nodes retained,while incorporating a semantic coverage term into the cost model to ensure thatwe retain trajectories which are semantically different. We demonstrate how ETScan achieve 1.8$\times$ reduction in average KV cache size during the searchprocess, leading to 1.4$\times$ increased throughput relative to priorstate-of-the-art methods, with minimal accuracy degradation and withoutrequiring any custom kernel implementation. Code is available at:https://github.com/SqueezeAILab/ETS.