High computation costs and latency of large language models such as GPT-4 have limited their deployment in clinical settings. Small language models (SLMs) offer a cost-effective alternative, but their limited capacity requires biomedical domain adaptation, which remains challenging. An additional bottleneck is the unavailability and high sensitivity of clinical data. To address these challenges, we propose a novel framework for adapting SLMs into high-performing clinical models. We introduce the MediPhi collection of 3.8B-parameter SLMs developed with our novel framework: pre-instruction tuning of experts on relevant medical and clinical corpora (PMC, Medical Guideline, MedWiki, etc.), model merging, and clinical-tasks alignment. To cover most clinical tasks, we extended the CLUE benchmark to CLUE+, doubling its size. Our expert models deliver relative improvements on this benchmark over the base model without any task-specific fine-tuning: 64.3% on medical entities, 49.5% on radiology reports, and 44% on ICD-10 coding (outperforming GPT-4-0125 by 14%). We unify the expert models into MediPhi via model merging, preserving gains across benchmarks. Furthermore, we built the MediFlow collection, a synthetic dataset of 2.5 million high-quality instructions on 14 medical NLP tasks, 98 fine-grained document types, and JSON format support. Alignment of MediPhi using supervised fine-tuning and direct preference optimization achieves further gains of 18.9% on average.

We investigate the computational limits of the memory retrieval dynamics of modern Hopfield models from the fine-grained complexity analysis. Our key contribution is the characterization of a phase transition behavior in the efficiency of all possible modern Hopfield models based on the norm of patterns. Specifically, we establish an upper bound criterion for the norm of input query patterns and memory patterns. Only below this criterion, sub-quadratic (efficient) variants of the modern Hopfield model exist, assuming the Strong Exponential Time Hypothesis (SETH). To showcase our theory, we provide a formal example of efficient constructions of modern Hopfield models using low-rank approximation when the efficient criterion holds. This includes a derivation of a lower bound on the computational time, scaling linearly with $\max$\{ \# of stored memory patterns, length of input query sequence\}. In addition, we prove its memory retrieval error bound and exponential memory capacity.

Medical doctors spend on average 52 to 102 minutes per day writing clinical notes from their patient encounters (Hripcsak et al., 2011).Reducing this workload calls for relevant and efficient summarization methods.In this paper, we introduce new resources and empirical investigations for the automatic summarization of doctor-patient conversations in a clinical setting.In particular, we introduce the MTS-DIALOG dataset; a new collection of 1,700 doctor-patient dialogues and corresponding clinical notes.We use this new dataset to investigate the feasibility of this task and the relevance of existing language models, data augmentation, and guided summarization techniques.We compare standard evaluation metrics based on n-gram matching, contextual embeddings, and Fact Extraction to assess the accuracy and the factual consistency of the generated summaries.To ground these results, we perform an expert-based evaluation using relevant natural language generation criteria and task-specific criteria such as critical omissions, and study the correlation between the automatic metrics and expert judgments.To the best of our knowledge, this study is the first attempt to introduce an open dataset of doctor-patient conversations and clinical notes, with detailed automated and manual evaluations of clinical note generation.

End-to-end network slicing is a promising concept based on softwarization and virtualization, leading the way towards efficiently achieving the network and operational key performance indicators (KPIs) for future wireless systems that comprise softwarized network functions. It leverages the underlying physical infrastructure to create and orchestrate agile and programmable network functions which satisfy the end-to-end user demands. These features are crucial during situations in which sudden demand surges stress the wireless system. Proper network orchestration can provide the necessary and timely adaptability to offer sustained communication between end-users. This paper addresses the design and implementation of an end-to-end network slicing framework specifically designed to provide orchestration tools for softwarized network functions, in order to fulfill the system requirements for surge events such as flash crowds. The conducted performance evaluations demonstrate the applicability of this approach and validate the proof-of-concept implementation.

Many modern day cloud services are composites of multiple smaller services working correctly together. This design has become increasingly prevalent due to the rise of the microservices application architecture, as well as service chaining in Network Function Virtualization (NFV). Future composite applications and services will be deployed on multi-tier clouds where their constituent microservices may be geographically spread over different regions. To optimize the delivery of such composites, the constituent microservices must be placed in locations where their clients, which may be other microservices, are able to meet certain QoS constraints. We propose an architecture and present a prototype system for incorporating network metrics into the auto-scaling and scheduling decisions of cloud management systems. Given a service with QoS constraints, our system monitors the network metrics (e.g. latency and bandwidth) of their clients. If a particular client is unable to receive the required latency or bandwidth of the service, our system auto-scales the service and strategically places the new instance(s) in a location capable of meeting the service quality, and re-directs traffic to the new instance.

The SAVI testbed is a multi-tier SDN-enabled heterogeneous cloud created as an application enablement platform for academic research. It was developed to investigate the hypothesis that the entire infrastructure can be virtualized and abstracted as a service. Key to this investigation was the creation of a resource management system based on the concept of Software-Defined Infrastructure (SDI), which enables integrated control and management over heterogeneous resources and exposes a programmable interface for users. In this paper, we present our experiences in designing and deploying the Canadian national SAVI testbed, and discuss the various lessons learned along the way. In addition, we describe our ongoing work to enhance the testbed management system to meet the demands and challenges of the IoT era.

The economies of scale afforded by cloud computing has been a driving force behind the rapid development and deployment of new cloud-based network applications and services. With the massive growth of IoT devices, we expect a sharp rise in the volume of traffic seen going to and coming from cloud datacenters, which will continue to grow over the next several years. Network Function Virtualization (NFV) is a recent concept which promises to grant network operators the required flexibility to quickly develop and provision new network functions and services in the cloud. As NFV is agnostic to the computing resource, we foresee scenarios where unconventional resources such as FPGAs and GPUs will be of benefit. To this end, we present an architecture based on Software-Defined Infrastructure (SDI) which offers an abstracted control and management interface over virtualized heterogeneous resources in the cloud. Through a unified set of APIs, this architecture enables both application developers and network operators to dynamically deploy and manage new services in the cloud alongside the underlying network that interconnects them, all in a fully software-defined manner. We demonstrate and evaluate an implementation of our NFV-enablement architecture using the SAVI testbed, a multi-tier and SDN-enabled cloud containing virtualized heterogeneous compute resources.

We present a framework for creating heterogeneous virtualized network function (VNF) service chains from cloud data center resources. Traditionally, these functions are packaged in software images within a catalog of networking applications that can be loaded onto a virtual machine CPU, and can be offered to users as a service. Our framework combines the best of both software and hardware by allowing users to chain traditional software-based VNFs with hardware-based VNFs that the user provides as an IP to generate a bitstream or a pre-generated VNF as part of a library. To accomplish this, our framework first creates the hardware bitstreams and programs the FPGA VNFs, loads any software VNFs requested, and programs the network to daisy chain the VNFs together. Furthermore, this enables an incremental design flow where the user can start by implementing a chain of VNFs in software and incrementally substitute software VNFs for their hardware counterparts. Our paper investigates two case studies to show the ability to switch between hardware and software VNFs in our framework and to demonstrate the benefit of using hardware VNFs. The first study is signature matching at fixed offsets, similar to matching packet headers. In this case study, the CPU can keep up at line-rate using specialized networking drivers. The second case study involves string matching within a packet, which requires scanning through the entire frame. In this case, the CPU performance drops to approximately 20 percent of the input rate, whereas the FPGA can continue to keep up at line-rate.

With the advent of SDN, efforts on network virtualization have accelerated. Many concepts have been developed to enable the coexistence of multiple logical network domains on a shared networking infrastructure while also ensuring isolation between them. In this context, we envision a multi-tenant cloud environment that grants users the ability to dynamically alter the behaviour of the underlying network, thus achieving a truer sense of Infrastructure as a service. In this paper, we propose a method for providing users in multi-tenant cloud environments open APIs for installing custom flows into the network at layer 2 (L2). Our proposed system ensures tenant isolation while detecting possible flow-level conflicts that can lead to tenancy violations and to problems in end-to-end reachability. We prototyped our system on a nationwide testbed environment, which showed that our proposed method reduces flow-level conflict detection time over existing proposals from ms to μs scale.

This paper introduces research that addresses the subjective assessment of Quality of Experience (QoE) during the entire life cycle of a video session. We define a video session life cycle as the time from when a user attempts to initiate playback, until such time that the video ends either from normal video conclusion or through a network-induced failure. We provide a detailed description of our assessment methodology designed to discern whether a user's QoE would be impacted by the presence of failures. To accomphsh this, we carefully select various test conditions to take into consideration the rating scale used, the types of impairments and failures seen by the user, and whether impaired videos are seen together with failed videos in multi-video sessions. The selection and creation of source video sequences are also discussed, as well as the use of between-subjects and within-subjects approaches for running our experiments in a controlled laboratory setting. Statistical analysis was carried out to interpret our experimental results. We compared the results of the between-subjects measures and the results of the within-subjects measures, and concluded that the introduction of a scale with an extended lower bound enabled subjects to more clearly express their dissatisfaction of videos with failures when compared to the traditional ITU 5-point rating scale. In addition, we observed that videos that were simply impaired but concluded normally did not have a statistically significant difference when an extended scale was used.

Programmability in wireless access networks can provide unprecedented flexibility in meeting the communications needs of a diverse set of wireless devices under changing demand and network conditions. Programmability also holds the promise of enabling multiple simultaneous virtual operators providing a variety of access networks that offer different services using different business models. In this paper we present our work on enabling wireless access network programmability in the SAVI testbed. We introduce an architecture for Software-Defined Infrastructure that offers virtualized heterogeneous resources in support of services and applications. Central to this architecture is the Janus SDI resource manager that can coordinate the actions of a set of controllers, e.g. OpenStack, OpenFlow, and other controllers such as a wireless access controller. We describe a design where Janus is used to integrate the wireless access network into a Smart Edge node. We introduce use cases that exploit the flexibility offered by this design, and we present experimental results from our implementation.

In this paper we discuss how to enable Software-Defined Networking (SDN) applications on Software-Defined Infrastructure (SDI) which is an approach for integrated control and management of converged computing and networking resources. Current separated resource management for computing or networking resources is not sufficient for addressing applications and multimedia services that require guaranteed service and quality levels. In addition, current resource management is not capable of managing heterogeneous resources that include computing and networking resources in combination with other resources such as programmable hardware, GPUs and network processors. We present an SDI that provides pluggable resource management modules for scheduling, networking control, fault management, and so on. This paper focuses on the design and implementation of a networking control module that enables SDN applications using information available from other modules. Currently, we have deployed the network control module in the practical multi-tier cloud infrastructure, SAVI Testbed. We present real measurements that show the functional evaluation results of our network control module.

Wireless applications in process automation started in areas where wireless sensors provide rich process information to the automation systems. Shortly after WirelessHART became the international standard, both ISA100.11a and WirelessHART claimed that their respective wireless technology could be applied to control as well. Although both standards provide provisions for supporting control over wireless, actual installations of wireless control systems have been slow to be adopted. Due to people's wariness of using wireless in combination with the serious nature of control, what is needed is a demonstration of performance and reliability of control when being executed across wireless media. A good starting point is a comparison between a wireless control system and its wired counterpart. In this paper we first study if we can establish a WirelessHART control loop the same way as a wired one; then we study if our WirelessHART control loop could achieve the same execution rates and reliability as a wired Foundation Fieldbus control loop. The experiment affirms that, in most likely cases, control over a WirelessHART network could be as good as control over a wired fieldbus.

Today's Web browsers allow users to open links in new windows or tabs. This action, which we call 'branching', is sometimes performed on search results when the user plans to eventually visit multiple results. We detect branching behavior on a large commercial search engine with a client-side script on the results page. Two-fifths of all users spawned new tabs on search results in the timeframe of our study; branching usage varied with different query types and vertical. Both branching and backtracking are viable methods for visiting multiple search results. To understand user search strategies, we treat multiple result clicks following a query as ordered events to understand user search strategies. Users branching in a query are more likely to click search results from top to bottom, while users who backtrack are less likely to do so; this is especially true for queries involving more than two clicks. These findings inform an experiment in which we take a popular click model and modify it to account for the differing user behavior when branching. By understanding that users continue examining search results before viewing a branched result, we can improve the click model for branching queries.

We introduce an entity-centric search experience, called Active Objects, in which entity-bearing queries are paired with actions that can be performed on the entities. For example, given a query for a specific flashlight, we aim to present actions such as reading reviews, watching demo videos, and finding the best price online. In an annotation study conducted over a random sample of user query sessions, we found that a large proportion of queries in query logs involve actions on entities, calling for an automatic approach to identifying relevant actions for entity-bearing queries. In this paper, we pose the problem of finding actions that can be performed on entities as the problem of probabilistic inference in a graphical model that captures how an entity bearing query is generated. We design models of increasing complexity that capture latent factors such as entity type and intended actions that determine how a user writes a query in a search box, and the URL that they click on. Given a large collection of real-world queries and clicks from a commercial search engine, the models are learned efficiently through maximum likelihood estimation using an EM algorithm. Given a new query, probabilistic inference enables recommendation of a set of pertinent actions and hosts. We propose an evaluation methodology for measuring the relevance of our recommended actions, and show empirical evidence of the quality and the diversity of the discovered actions.

How can we cull the facts we need from the overwhelming mass of information and misinformation that is the Web? The TextRunner extraction engine represents one approach, in which people pose keyword queries or simple questions and TextRunner returns concise answers based on tuples extracted from Web text. Unfortunately, the results returned by engines such as TextRunner include both informative facts (e.g., “the FDA banned ephedra”) and less useful statements (e.g., “the FDA banned products”). This paper therefore investigates filtering TextRunner results to enable people to better focus on interesting assertions. We first develop three distinct models of what assertions are likely to be interesting in response to a query. We then fully operationalize each of these models as a filter over TextRunner results. Finally, we develop a more sophisticated filter that combines the different models using relevance feedback. In a study of human ratings of the interestingness of TextRunner assertions, we show that our approach substantially enhances the quality of TextRunner results. Our best filter raises the fraction of interesting results in the top thirty from 41.6 % to 64.1%.