Agents based on large language models (LLMs) for machine learning engineering (MLE) can automatically implement ML models via code generation. However, existing approaches to build such agents often rely heavily on inherent LLM knowledge and employ coarse exploration strategies that modify the entire code structure at once. This limits their ability to select effective task-specific models and perform deep exploration within specific components, such as experimenting extensively with feature engineering options. To overcome these, we propose MLE-STAR, a novel approach to build MLE agents. MLE-STAR first leverages external knowledge by using a search engine to retrieve effective models from the web, forming an initial solution, then iteratively refines it by exploring various strategies targeting specific ML components. This exploration is guided by ablation studies analyzing the impact of individual code blocks. Furthermore, we introduce a novel ensembling method using an effective strategy suggested by MLE-STAR. Our experimental results show that MLE-STAR achieves medals in 64% of the Kaggle competitions on the MLE-bench, significantly outperforming the best alternative.

Service meshes have become essential for enabling microservices in cloud environments; however, they also introduce substantial network overhead. In particular, the ingress gateway, which serves as the primary entry point for external traffic, has emerged as a major performance bottleneck due to CPU-intensive traffic analysis and prolonged forwarding paths through multiple network stack layers. Our analysis indicates that these inefficiencies can result in a 4-fold reduction in network throughput and increased CPU resource consumption. In response, we propose HardMesh, a hardware-software hybrid ingress gateway that leverages a Smart-NIC for high-performance traffic analysis and efficient traffic routing. This process is augmented by a lightweight CPU-based proxy for traffic management. Evaluations show that HardMesh outperforms existing ingress gateways, achieving up to 4.4× higher throughput while providing the same range of traffic management services.

With the increasing popularity of containers for deploying microservices, ensuring the security of container networks has become a vital concern. However, current security solutions rely on a host's operating system (OS) to enforce network policies for container traffic. This design incurs severe overhead and cannot guarantee container network security when attackers gain access to the host's OS. Therefore, we propose HardWhale, a hardware-isolated network security enforcement system for containers that delivers high-performance and robust network security without depending on the host's OS. HardWhale leverages a smartNIC, physically isolating the entire container traffic inspection stack from the host and accelerating inspection tasks. Inspection policies securely reside within the smartNIC and are updated in runtime without involving the host, due to our isolated policy management mechanism. This design ensures robust network security for containers, even if the host is exposed to attackers. Evaluations show that HardWhale protects containers against various network attacks in compromised environments and improves HTTP throughput threefold and HTTP latency 2.3-fold compared to state-of-the-art solutions.

Large language models (LLMs) have made significant advancements in various natural language processing tasks, including question answering (QA) tasks. While incorporating new information with the retrieval of relevant passages is a promising way to improve QA with LLMs, the existing methods often require additional fine-tuning which becomes infeasible with recent LLMs. Augmenting retrieved passages via prompting has the potential to address this limitation, but this direction has been limitedly explored. To this end, we design a simple yet effective framework to enhance open-domain QA (ODQA) with LLMs, based on the summarized retrieval (SuRe). SuRe helps LLMs predict more accurate answers for a given question, which are well-supported by the summarized retrieval that could be viewed as an explicit rationale extracted from the retrieved passages. Specifically, SuRe first constructs summaries of the retrieved passages for each of the multiple answer candidates. Then, SuRe confirms the most plausible answer from the candidate set by evaluating the validity and ranking of the generated summaries. Experimental results on diverse ODQA benchmarks demonstrate the superiority of SuRe, with improvements of up to 4.6\% in exact match (EM) and 4.0\% in F1 score over standard prompting approaches. SuRe also can be integrated with a broad range of retrieval methods and LLMs. Finally, the generated summaries from SuRe show additional advantages to measure the importance of retrieved passages and serve as more preferred rationales by models and humans.

Developing an effective molecular generation framework even with a limited number of molecules is often important for its practical deployment, e.g., drug discovery, since acquiring task-related molecular data requires expensive and time-consuming experimental costs. To tackle this issue, we introduce Hierarchical Textual Inversion for Molecular Generation (HI-Mol), a novel data-efficient molecular generation method. HI-Mol is inspired by the importance of hierarchical information, e.g., both coarse- and fine-grained features, in understanding the molecule distribution. We propose to use multi-level embeddings to reflect such hierarchical features based on the adoption of the recent textual inversion technique in the visual domain, which achieves data-efficient image generation. Compared to the conventional textual inversion method in the image domain using a single-level token embedding, our multi-level token embeddings allow the model to effectively learn the underlying low-shot molecule distribution. We then generate molecules based on the interpolation of the multi-level token embeddings. Extensive experiments demonstrate the superiority of HI-Mol with notable data-efficiency. For instance, on QM9, HI-Mol outperforms the prior state-of-the-art method with 50x less training data. We also show the effectiveness of molecules generated by HI-Mol in low-shot molecular property prediction. Code is available at https://github.com/Seojin-Kim/HI-Mol.

In tabular prediction tasks, tree-based models combined with automated feature engineering methods often outperform deep learning approaches that rely on learned representations. While these feature engineering techniques are effective, they typically depend on a pre-defined search space and primarily use validation scores for feature selection, thereby missing valuable insights from previous experiments. To address these limitations, we propose a novel tabular learning framework that utilizes large language models (LLMs), termed Optimizing Column feature generator with decision Tree reasoning (OCTree). Our key idea is to leverage the reasoning capabilities of LLMs to identify effective feature generation rules without manually specifying the search space and provide language-based reasoning information highlighting past experiments as feedback for iterative rule improvements. We use decision trees to convey this reasoning information, as they can be easily represented in natural language, effectively providing knowledge from prior experiments (i.e., the impact of the generated features on performance) to the LLMs. Our empirical results demonstrate that OCTree consistently enhances the performance of various prediction models across diverse benchmarks, outperforming competing automated feature engineering methods. Code is available at https://github.com/jaehyun513/OCTree.

Containers have become the predominant virtualization technique for deploying microservices in cloud environments. However, container networking, critical for microservice functionality, often introduces significant overhead and resource consumption, potentially degrading the performance of microservices. This challenge arises from the complexity of the software-based network data plane, responsible for network virtualization and access control within container traffic. To tackle this challenge, we proposeHyperion, a novel hardware-based container networking system that prioritizes high performance and security. Leveraging smartNICs, commonly found in cloud environments,Hyperionimplements a fully-functional container network data plane, encompassing network virtualization and access control. It also has the capability to dynamically optimize its data plane for agile responses to frequent changes in container environments, ensuring up-to-date data plane operation. This hardware-based design empowersHyperionto significantly improve the overall container networking performance without relying on the host system resources. Notably,Hyperionseamlessly integrates with existing containerized applications without necessitating modifications. Our evaluation shows that compared to state-of-the-art solutions,Hyperionachieves significant improvements in HTTP container communication latency and throughput by up to 2.25x and 4.3x, respectively. Furthermore, it reduces CPU utilization associated with container networking by up to 4x.

Distributed SDN (Software-Defined Networking) controllers have rapidly become an integral element ofWide Area Networks (WAN), particularly within SD-WAN, providing scalability and fault-tolerance for expansive network infrastructures. However, the architecture of these controllers introduces new potential attack surfaces that have thus far received inadequate attention. In response to these concerns, we introduceAmbusher, a testing tool designed to discover vulnerabilities within protocols used in distributed SDN controllers.Ambusherachieves this by leveragingprotocol state fuzzing, which systematically finds attack scenarios based on an inferred state machine. Since learning states from a cluster is complicated,Ambusherproposes a novel methodology that extracts a single and relatively simple state machine, achieving efficient state-based fuzzing. Our evaluation ofAmbusher, conducted on a real SD-WAN deployment spanning two campus networks and one enterprise network, illustrates its ability to uncover 6 potential vulnerabilities in the widely used distributed controller platform.

With the increasing adoption of containerization in cloud services, container networking has become a critical concern, as it enables the agile deployment of microservices but also introduces new vulnerabilities susceptible to network attacks, posing a threat to container environments. While several security solutions have been introduced to address this concern, they unfortunately exhibit significant shortcomings, including security vulnerabilities and limited performance. We thus propose Helios, a novel hardware-based network security extension that addresses the security and performance limitations in existing solutions. Leveraging a smartNIC, Helios enhances both the security and performance facets of container networking through two key mechanisms: (i) the establishment of physically isolated container communication channels and (ii) the network security engines fully offloaded to the smartNIC. Our evaluation shows that Helios mitigates various network threats initiated from both container- and host-side while performing up to 3x faster than the existing solutions in container communication.

While the use of containerization technologies for virtual application deployment has grown at an astonishing rate, the question of the robustness of container networking has not been well scrutinized from a security perspective, even though inter-container networking is indispensable for microservices. Thus, this paper first analyzes container networks from a security perspective, discussing the implications based on their architectural limitations. Then, it presents Bastion+, a secure inter-container communication bridge. Bastion+ introduces ($i$) a network security enforcement stack that provides fine-grained control per container application and securely isolates inter- container traffic in a point-to-point manner. Bastion+ also supports ($ii$) selective security function chaining, enabling various security functions to be chained between containers for further security inspections (e.g., deep packet inspection) according to the container’s network context. Bastion+ incorporates ($iii$) a security policy assistant that helps an administrator discover inter-container networking dependencies correctly. Our evaluation demonstrates how Bastion+ can effectively mitigate several adversarial attacks in container networks while improving the overall performance up to 25.4% within single-host containers and 17.7% for cross-host container communications.

Network virtualization (NV) has been widely used today in data centers to meet the multi-tenancy requirement-a key to achieving Infrastructure as a Service (IaaS) in a modern cloud environment. However, even though NV is a popular solution employed by network operators, we argue that most solutions still have challenges in terms of (i) configuration, (ii) management, and (iii) security; all of which hinder the deployment of secure and practical virtual networks that tenants wish to construct. To bridge the gaps, we propose LinkWire, a new NV system that produces secure and reconfigurable virtual networks. For this, LinkWire leverages extended Berkeley Packet Filter (eBPF), the recently-adopted in-kernel programmable networking technology. We also illustrate several use cases to show how our system brings benefits.

Software-Defined Networking (SDN) has aggressively grown in data center networks, telecommunication providers, and enterprises by virtue of its programmable and extensible control plane. Also, there have been many kinds of research on the security of SDN components along with the growth of SDN. Some of them have inspected network policy inconsistency problems that can severely cause network reliability and security issues in SDN. However, they do not consider whether a single network policy itself is corrupted during processing inside and between SDN components. In this paper, we thus focus on the question of how to automatically identify cases in which the SDN stack fails to prevent policy inconsistencies from arising among those components. We then present AudiSDN, an automated fuzz-testing framework designed to formulate test cases in which policy inconsistencies can arise inOpenFlownetworks, the most prevalent SDN protocol. To prove its feasibility, we applied AudiSDN to two widely used SDN controllers, Floodlight and ONOS, and uncovered three separate CVEs (Common Vulnerabilities and Exposures) that cause the network policy inconsistencies among SDN components. Furthermore, we investigate the design flaws that cause the inconsistencies in modern SDN components, suggesting specific validations to address such a serious but understudied pragmatic concern.

Regular expression (regex) matching is an integral part of deep packet inspection (DPI) but a major bottleneck due to its low performance. For regex matching (REM) acceleration, FPGA-based studies have emerged and exploited parallelism by matching multiple regex patterns concurrently. However, even though guaranteeing high-performance, existing FPGA-based regex solutions do not still support dynamic updates in run time. Hence, it was inappropriate as a DPI function due to frequently altered malicious signatures. In this work, we introduce Reinhardt, a real-time reconfigurable hardware architecture for REM. Reinhardt represents regex patterns as a combination of reconfigurable cells in hardware and updates regex patterns in real-time while providing high performance. We implement the prototype using NetFPGA-SUME, and our evaluation demonstrates that Reinhardt updates hundreds of patterns within a second and achieves up to 10 Gbps throughput (max. hardware bandwidth). Our case studies show that Reinhardt can operate as NIDS/NIPS and as the REM accelerator for them.

The robustness of a network’s connectivity to other networks is often highly dependent on a few critical nodes and links that tie the network to the larger topology. The failure or degradation to such network bottlenecks can result in outages that may propagate throughout the network. Unfortunately, the presence of the bottlenecks also offers opportunities for targetedlink flooding attacks (LFAs). Researchers have proposed a new and promising defense to counter LFAs, referred to astopology deception. This strategy centers on hindering the discovery of bottlenecks by presenting false trace responses to adversaries as they perform topological probing of the target network. Even though the goal of topology deception centers on obscuring critical links, node dependencies can be exploited by an adversary. However, current approaches do not consider a wide range of metrics that may reveal important and diverse aspects of network bottlenecks. Furthermore, existing approaches create a simple form of virtual topology, which is subject to relatively easy detection by the adversary, reducing its effectiveness. In this paper, we propose a comprehensive topology deception framework, which we refer to as BottleNet. Our suggested approach can analyze various network topology features both with respect to static and dynamic metrics and then use this information to identify bottlenecks, finally producing complex virtual topologies that are resilient to adversarial detection.

While containerization has emerged as a lightweight approach to package, deploy, and run legacy applications in a resource-efficient manner, the shared kernel-resource model used by containers introduces critical security concerns. Specifically, the abuse of system calls by a compromised container can trigger the security vulnerabilities of a host kernel. Unfortunately, even though existing solutions provide powerful protection mechanisms against such issues, how to define the capabilities of containers is still up to operators. In this work, we thus introduce TCLP, a dynamic analysis system that helps operators configure the least capabilities of containers to protect not only themselves but also a host. TCLP monitors the system calls triggered by containers in run time and finds the least capabilities required to run the containers based on the collected system calls. Finally, operators configure the minimal capabilities discovered by TCLP for their containers, reducing the risk of kernel vulnerabilities.

Barista is a novel architecture that seeks to enable flexible and customizable instantiations of network operating systems (NOSs) for software-defined networks (SDNs). As the NOS is the strategic control center of an SDN, implementing logic for management of network switches as well as higher-level applications, its design is critical to the welfare of the network. In this paper, we focus on three aspects of composable controller design: component synthesis, dynamic event control, and predictive NOS assessment. First, the modular design of the Barista enables flexible composition of functionalities prevalent in contemporary SDN controllers. Second, its event handling mechanism enables dynamic customization of control flows in a NOS. Third, its predictive NOS assessment helps to discover the optimal composition for the requirements specified by operators. These capabilities allow Barista operators to optimally select functionalities and dynamically handle events for their operating requirements while maximizing the resource utilization of the given system. Our results demonstrate that Barista can synthesize NOSs with many functionalities found in commodity controllers with competitive performance profiles.

As the network operating system (NOS) is the strategic control center of a software-defined network (SDN), its design is critical to the welfare of the network. Contemporary research has largely focused on specialized NOSs that seek to optimize controller design across one or a few dimensions (e.g., scalability, performance, or security) due to fundamental differences in architectural trade-offs needed to support competing demands. We thus designed Barista, as a new framework that enables flexible and customizable instantiations of network operating systems (NOSs) supporting diverse design choices. The Barista framework incorporates two mechanisms to harmonize architectural differences across design choices: component synthesis and dynamic event control. First, the modular design of the Barista framework enables flexible composition of functionalities prevalent in contemporary SDN controllers. Second, its event-handling mechanism enables dynamic adjustment of control flows in a NOS. These capabilities allow operators to easily enable functionalities and dynamically handle associated events, thereby satisfying network operating requirements. Our results demonstrate that Barista can synthesize NOSs with many functionalities found in commodity NOSs with competitive performance profiles.

While cloud computing platform becomes popular and works as a platform for network function virtualization (NFV), the security of the cloud also becomes an important subject. However, although there are many works about security mechanisms, there has not been much research into what problems can occur when these conventional mechanisms are applied to the cloud system. Thus, we have given more attention to the robustness of communications resided in the cloud, not security mechanism itself, and found that security threats could arise from communication between cloud services and identification process. To cope with this problem, we propose three approaches: integrative identification system in a single cloud service, action-based token authorization, and partially encrypted communication between the identification system and cloud services. By implementing these approaches to open-source cloud computing platform, Openstack, we show that our approaches are feasible.

Software-Defined Networking (SDN), controlling underlying network devices (i.e., data plane) in a logically centralized manner, is now actively adopted in many real world networking environments. It is clear that a network administrator can easily understand and manage his networking environments with the help of SDN. In SDN, a network operating system (NOS), also known as an SDN controller, is the most critical component because it should be involved in all transactions for controlling network devices, and thus the security of NOS cannot be highly exaggerated. However, in spite of its importance, no previous works have thoroughly investigated the security of NOS. In this work, to address this problem, we present the NOSArmor, which integrates several security mechanisms, named as security building block (SBB), into a consolidated SDN controller. NOSArmor consists of eight SBBs and each of them addresses different security principles of network assets. For example, while role-based authorization focuses on securing confidentiality of internal storage from malicious applications, OpenFlow protocol verifier protects availability of core service in the controller from malformed control messages received from switches. In addition, NOSArmor shows competitive performance compared to existing other controllers (i.e., ONOS, Floodlight) with secureness of network assets.

We design Barista, as a new framework that seeks to enable flexible and customizable instantiations of network operating systems (NOSs) supporting diverse design choices, using two key features that harmonize architectural differences across design choices: component synthesis and dynamic event control. With these capabilities, Barista operators to easily enable functionalities and dynamically adjust the control flows among those functionalities.

In this work, we present a highly scalable network intrusion detection system on many-core processors. To maximize the NIDS performance, we take advantage of the underlying hardware and adhere to four design principles: shared-nothing architecture, computation offloading, lightweight data structure, and flow offloading. Through the experimental results, we find that our design choices can significantly improve the NIDS performance (79 Gbps with 1514B synthetic packets). We believe that our design decisions can be easily extended to other many-core processors and programmable NICs.