Cisco College

Based on a pre-sales tech project customer churn analysis of 5,500 interactions, we identified three important drivers of churn: proposal presentation, participation in the solution showcase, and C-level stakeholder involvement. Machine learning models (XGBoost, Random Forest, Support Vector Machines) correctly predicted churn at <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 6. 5 \%}$</tex> to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 2. 3 \%}$</tex>. Demonstration of power and stakeholder engagement quality were associated with successful conversions. By combining measures of engagement, stakeholder profiles, and competitive strength, a predictive model emerged to tackle pre-sales churn risk. The work complements literature that has focused on postpurchase churn, and it gives sales organizations superior resource allocation, earlier risk detection, and better conversion. The work demonstrates pre-sales behavior can be correctly modeled and produces actionable recommendations to optimize prospect retention strategies in complex tech selling.

Quantum computing continues to face challenges in scaling both qubit quality and quantity, motivating the shift toward Distributed Quantum Computing (DQC), where multiple interconnected Quantum Processing Units (QPUs) collaboratively solve problems too complex for monolithic systems. In a quantum data center, QPUs distributed across racks can perform remote gate operations by creating entanglement between them, enabling circuit execution through DQC. However, generating end-to-end entangled pairs in a quantum data center often causes congestion and resource contention. To mitigate this, we introduce a resource management framework with efficient schedulers that maximize fidelity-guaranteed throughput while preserving the dependency constraints of sub-circuits derived from the original quantum circuit. We benchmark the scheduling problem using a MixedInteger Linear Programming (MILP) model and then propose a set of lightweight schedulers-Least Resource First (LRF) and its extension, Group LRF (GLRF). These lightweight schedulers provide practical, though less optimal, solutions for handling the high volume of remote gates in DQC workloads. Our initial results on a DCell-based quantum data center show that such heuristics perform well under current DQC conditions, balancing runtime and performance while avoiding the heavy overhead of solvers or machine learning approaches, which often struggle to meet the strict timing demands of short-lived entanglement.

Kubernetes is a widely used platform for deploying and managing containerized applications due to its efficient elastic capabilities. The Horizontal Pod Autoscaler (HPA) in Kubernetes independently adjusts the number of pods for each service, yet these services often operate in an interconnected manner. This study aims to understand the effects of autoscaling events on a graph of interconnected services. To achieve this, we apply control theory to model the HPA’s behavior. We analyze the stability of this model, perform numerical simulations, and deploy a real testbed to evaluate the performance. Our findings demonstrate that the control theory-based model accurately predicts the HPA’s behavior, ensuring system stability with CPU utilization meeting desired thresholds and no traffic loss after a transitional period. The model provides insights into optimizing resource scheduling and improving application performance in Kubernetes environments. Additionally, we extend our model to the whole service graph to understand how individual scaling decisions influence the complex graphs of cloud applications.

The single instruction multiple data (SIMD) capability in modern processors is critical to improving the performance of current compute-intensive programs. Modern compilers use vectorization techniques to exploit the SIMD capability, by detecting data parallelism in scalar source code and transforming a group of scalar instructions into vector-based instructions. In this study, we focus on one of the most common vectorization techniques, a technique called loop-based vectorization, which targets loops and optimizes their performance by grouping multiple occurrences of the same operation across loop iterations into a single SIMD instruction. We propose a data-driven graph-based learning framework for automatic vectorization, called autograph , which takes an input program, extracts the loops, and then learns a structured representation to automatically predict the correct vectorization and interleaving factors. Our proposed framework utilizes deep reinforcement learning to learn an optimal policy (observations to actions) from an intelligent agent in a SIMD environment, and automatically injects the predicted vectorization pragmas into the input program. We conducted an extensive evaluation on multiple benchmark datasets and comparisons with state-of-the-art baselines. Our results show that autograph achieves on average 2.49× performance improvement for Polybench compared to NeuroVectorizer and 3.69× compared to the baseline -O3.

Hierarchical Federated Learning (HFL) has shown great promise over the past few years, with significant improvements in communication efficiency and overall performance. However, current research for HFL predominantly centers on supervised learning. This focus becomes problematic when dealing with semi-supervised learning, particularly under non-IID scenarios. In order to address this gap, our paper critically assesses the performance of straightforward adaptations of current state-of-the-art semi-supervised FL (SSFL) techniques within the HFL framework. We also introduce a novel clustering mechanism for hierarchical embeddings to alleviate the challenges introduced by semi-supervised paradigms in a hierarchical setting. Our approach not only provides superior accuracy, but also converges up to 5.11× faster, while being robust to non-IID data distributions for multiple datasets with negligible communication overhead.<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup>

Silicon Photonics has emerged as the leading technology platform for high-speed, high density, and low power optical interconnects addressing hyperscale datacenter, high-performance computing, mobile and enterprise applications. Advanced 3D integration using through silicon vías (TSVs) for low parasitic interconnection and power delivery, along with high throughput, submicron waveguide to optical fiber attachment and laser integration will be essential to enable higher data rate per lane interconnects in Tbps optical transceivers on the horizon. In this paper, we will review trends in photonics packaging architecture for scale up and scale out of optical interconnects, critical technology challenges and path to technology adoption for high volume productization.

We present Mirror, an open-source platform for data exploration and analysis powered by large language models. Mirror offers an intuitive natural language interface for querying databases, and automatically generates executable SQL commands to retrieve relevant data and summarize it in natural language. In addition, users can preview and manually edit the generated SQL commands to ensure the accuracy of their queries. Mirror also generates visualizations to facilitate understanding of the data. Designed with flexibility and human input in mind, Mirror is suitable for both experienced data analysts and non-technical professionals looking to gain insights from their data.1