Collaborative training methods like Federated Learning (FL) and Split Learning (SL) enable distributed machine learning without sharing raw data. However, FL assumes clients can train entire models, which is infeasible for large-scale models. In contrast, while SL alleviates the client memory constraint in FL by offloading most training to the server, it increases network latency due to its sequential nature. Other methods address the conundrum by using local loss functions for parallel client-side training to improve efficiency, but they lack server feedback and potentially suffer poor accuracy. We propose FSL-SAGE (Federated Split Learning via Smashed Activation Gradient Estimation), a new federated split learning algorithm that estimates server-side gradient feedback via auxiliary models. These auxiliary models periodically adapt to emulate server behavior on local datasets. We show that FSL-SAGE achieves a convergence rate of $\mathcal{O}(1/\sqrt{T})$, where $T$ is the number of communication rounds. This result matches FedAvg, while significantly reducing communication costs and client memory requirements. Our empirical results also verify that it outperforms existing state-of-the-art FSL methods, offering both communication efficiency and accuracy.

Simulation models of law-abiding pedestrians are all too prevalent. However, the models of dangerous pedestrians are frighteningly limited. This leads to the unreliability of autonomous driving in human habitats as they learn to deal with pedestrians exhaustively in a simulated world. There exists a shortage of data and analysis on dangerous pedestrians. Decision-making factors and behavior identification are not sufficient; a dangerous pedestrian model should consist of a collection of behaviors that represent their natural behavior pattern. On this basis, we propose to define these collections of behaviors in the form of pedestrian archetypes. Each archetype suggests a distinct pedestrian personality with their own special take on crossing the road, giving autonomous driving an opportunity to significantly improve their testing strategies and reliability against pedestrians.

Agentic search such as Deep Research systems-where agents autonomously browse the web, synthesize information, and return comprehensive citation-backed answers-represents a major shift in how users interact with web-scale information. While promising greater efficiency and cognitive offloading, the growing complexity and open-endedness of agentic search have outpaced existing evaluation benchmarks and methodologies, which largely assume short search horizons and static answers. In this paper, we introduce Mind2Web 2, a benchmark of 130 realistic, high-quality, and long-horizon tasks that require real-time web browsing and extensive information synthesis, constructed with over 1000 hours of human labor. To address the challenge of evaluating time-varying and complex answers, we propose a novel Agent-as-a-Judge framework. Our method constructs task-specific judge agents based on a tree-structured rubric design to automatically assess both answer correctness and source attribution. We conduct a comprehensive evaluation of ten frontier agentic search systems and human performance, along with a detailed error analysis to draw insights for future development. The best-performing system, OpenAI Deep Research, can already achieve 50-70% of human performance while spending half the time, highlighting its great potential. Altogether, Mind2Web 2 provides a rigorous foundation for developing and benchmarking the next generation of agentic search systems.

Densely-deployed femtocell networks are used to enhance wireless coverage in public spaces such as office buildings, subways, and academic buildings. These networks can increase user throughput, but edge users can suffer from co-channel interference and service outages. This paper introduces a distributed algorithm for network configuration, called Radius Reduction and Scheduling (RRS), to improve the performance and fairness of the network. RRS works by jointly adapting femtocell transmission power, allocating user to femtocells, and scheduling resource blocks so as to increase fairness and reduce outage probability in dense femtocell networks. RRS produces a network configuration that guarantees either user or area coverage depending on the management needs. A prototype implementation confirms the benefits of RRS in a real environment. Furthermore, extensive simulations show that RRS reduces the outage probability of up to 50%, and provides better fairness, with an increase in Jain's index of 190%, with respect to a baseline algorithm which works with fixed power and best-effort scheduling and to a previous approach to resource management in femtocell networks.

Densely-deployed femtocell networks are used to enhance wireless coverage in public spaces like office buildings, subways, and academic buildings. These networks can increase throughput for users, but edge users can suffer from co-channel interference, leading to service outages. This paper introduces a distributed algorithm for network configuration, called Radius Reduction and Scheduling (RRS), to improve the performance and fairness of the network. RRS determines cell sizes using a Voronoi-Laguerre framework, then schedules users using a scheduling algorithm that includes vacancy requests to increase fairness in dense femtocell networks. We prove that our algorithm always terminate in a finite time, producing a configuration that guarantees user or area coverage. Simulation results show a decrease in outage probability of up to 50%, as well as an increase in Jain's fairness index of almost 200%.

Over-provisioned femtocell networks can be used to serve indoor locations that see high peak loads, such as airports or train stations. However, networks designed for high peak loads are mostly under-utilized, which is wasteful from an energy-use perspective. This paper introduces a femtocell selective activation problem. We motivate the use of selective activation in femtocell networks using real femtocell power measurements. We formally define the selective activation problem, and introduce GreenFemto, a distributed femtocell selective activation algorithm. We prove that GreenFemto converges to a locally Pareto optimal solution. Detailed simulations of an LTE wireless system are used to demonstrate the performance of GreenFemto. We find that GreenFemto uses up to 55% fewer femtocells to serve a given load, relative to an existing femtocell power-saving technique. Furthermore, we show that GreenFemto comes within 15% of a globally optimal solution. We conclude that selective activation can be successfully applied to femtocell networks to both reduce power consumption, and reduce outage probabilities.

Provisioning the capacity of wireless networks is difficult when peak load is significantly higher than average load, for example, in public spaces like airports or train stations. Service providers can use femtocells and small cells to increase local capacity, but deploying enough femtocells to serve peak loads requires a large number of femtocells that will remain idle most of the time, which wastes a significant amount of power. To reduce the energy consumption of over-provisioned femtocell networks, we formulate a femtocell selective activation problem, which we formalize as an integer nonlinear optimization problem. Then we introduce Green Femto, a distributed femtocell selective activation algorithm that deactivates idle femtocells to save power and activates them on-the-fly as the number of users increases. We prove that Green Femto converges to a locally Pareto optimal solution and demonstrate its performance using extensive simulations of an LTE wireless system. Overall, we find that Green Femto requires up to 55% fewer femtocells to serve a given user load, relative to an existing femtocell power-saving procedure, and comes within 15% of a globally optimal solution.

Large-scale enterprise femtocell deployments can significantly impact the performance and energy consumption of the underlying wireless networks that support them. Naive femtocell deployments can lead to higher overall network energy usage, while optimized femtocell deployments can increase total network connectivity and reduce macrocell energy consumption. This paper approaches femtocell deployment as a combinatorial optimization problem. We first consider accelerated greedy algorithms using one of two metrics: femtocell coverage and area spectral efficiency. Then, motivated by an analysis of the strengths and weaknesses of each metric, we introduce an algorithm that takes the weighted sum of both metrics. We evaluate our algorithms using extensive simulations, and find that our weighted sum algorithm decreases outage probability by up to 30% relative to existing greedy approaches. Furthermore, our algorithm can lead to a reduction in total network energy usage by up to 14%.

Enterprise femtocell deployments present numerous design challenges for network operators due to their high density, poor radio environments, and mixture of public and private users. Among these challenges, femtocell pilot management is particularly important due to its potential impact on both femtocell and macrocell users. Femtocell pilots determine femtocell coverage, but must be designed with consideration for macrocell transmissions and users. We introduce a utility-based femtocell pilot management algorithm that sets femtocell pilots by finding the Nash equilibrium of an N-player strategic game, using a utility function derived from the femtocell coverage radius. Using simulations, we find that the game-theoretic interference management algorithm reduces outage probabilities relative to a naive interference management scheme by up to 43% in the open access femtocell case, and 17% in the closed access femtocell case. Our conclusion is that game theory can be applied to the distributed problem of femtocell pilot estimation, with good results.

Femtocells are tiny, low-cost, customer-deployed cellular base stations that are designed to provide supplementary coverage to existing wireless networks. However, naive femtocell deployments can cause pilot pollution and raise interference levels for users connected to macrocells, particularly when the local density of femtocells is very high, such as in an enterprise setting. We describe three dynamic femtocell pilot adjustment algorithms that balance femtocell coverage and interference: a distributed and two cooperative algorithms that leverage a lightweight, local, interference management server. Using simulations, we find that the interference management algorithms reduce outage probabilities relative to a naive interference management scheme by up to 85%. Exploiting locality through the use of dynamic interference management algorithms is a promising approach to managing femtocell deployments.

3G wireless data networks have emerged as the first high speed, ubiquitous data networks. 4G wireless networks are poised to replace 3G networks as the next generation of mobile data networks, but the transition will be lengthy and expensive. Wireless authentication across 3G and 4G data networks is typically an arduous task, requiring complete disconnection from the existing network before performing a lengthy full authentication on the new network. We introduce an efficient interworking method for CDMA2000 1xEVDO and WiMAX data networks that allows existing authentication credentials to be leveraged in more efficient interworking protocols at both tightly- and loosely-coupled network integration levels, as well as a proactive handoff scheme that takes advantage of our tightly-coupled interworking. We perform a detailed simulation and mathematical analysis of our interworking and handoff protocols and find that our interworking scheme reduces interworking delays by up to 75% in the WiMAX to CDMA case and 85% in the CDMA to WiMAX case.

OBJECTIVE: To formulate a model for translating manual infection control surveillance methods to automated, algorithmic approaches. DESIGN: We propose a model for creating electronic surveillance algorithms by translating existing manual surveillance practices into automated electronic methods. Our model suggests that three dimensions of expert knowledge be consulted: clinical, surveillance, and informatics. Once collected, knowledge should be applied through a process of conceptualization, synthesis, programming, and testing. RESULTS: We applied our framework to central vascular catheter associated bloodstream infection surveillance, a major healthcare performance outcome measure. We found that despite major barriers such as differences in availability of structured data, in types of databases used and in semantic representation of clinical terms, bloodstream infection detection algorithms could be deployed at four very diverse medical centers. CONCLUSIONS: We present a framework that translates existing practice-manual infection detection-to an automated process for surveillance. Our experience details barriers and solutions discovered during development of electronic surveillance for central vascular catheter associated bloodstream infections at four hospitals in a variety of data environments. Moving electronic surveillance to the next level-availability at a majority of acute care hospitals nationwide-would be hastened by the incorporation of necessary data elements, vocabularies and standards into commercially available electronic health records.

The vast expansion of interconnectivity with the Internet and the rapid evolution of highly-capable but largely insecure mobile devices threatens cellular networks. In this paper, we characterize the impact of the large scale compromise and coordination of mobile phones in attacks against the core of these networks. Through a combination of measurement, simulation and analysis, we demonstrate the ability of a botnet composed of as few as 11,750 compromised mobile phones to degrade service to area-code sized regions by 93%. As such attacks are accomplished through the execution of network service requests and not a constant stream of phone calls, users are unlikely to be aware of their occurrence. We then investigate a number of significant network bottlenecks, their impact on the density of compromised nodes per base station and how they can be avoided. We conclude by discussing a number of countermeasures that may help to partially mitigate the threats posed by such attacks.

The widespread acceptance of RFID tags in inventory control applications has allowed the development of increasingly complex and useful inventory management techniques. This paper approaches the problem of real-time inventory querying in a large warehouse. We use a combination of powered, wireless- capable active RFID tags and a mobile RFID reader to design a querying system that uses both a mesh network formed with the active RFID tags and the mobile reader to balance query latency and total network lifetime. We implemented and simulated our system on Crossbow MicaZ motes and a custom robot platform. Our results show that our hybrid algorithm balances query latency and network lifetime more effectively than mesh network- only and mobile reader-only algorithms. In particular, we see network power consumption reduction of up to 60% and a 15% reduction in total distance travelled by the mobile reader. We conclude that introducing a mobile reader to active tag networks allows the design of flexible network algorithms that can reduce the impact of battery life on the network.