PRODUCT MONOGRAPH PrOTEZLA® 10 mg, 20 mg, and 30 mg tablets Selective Immunosuppressant Date of Preparation: 6755 Mississauga Road November 12, 2014 Submission Control No: 169862 © 2014 Celgene Corporation. ® OTEZLA is a registered trademark of Celgene Corporation. Page 1 of 28 TABLE OF CONTENTS Page 2 of 28 OTEZLA®
Report on research visit to University of California, Irvine (UCI) under HEQEP CP 3137 Dr. Md Yusuf Sarwar Uddin Deputy SPM CP 3137 Assistant Professor Department of CSE, BUET, Dhaka-1000.
I made a research visit to University of California, Irvine (UCI) United States fromSeptember 20, 2014 to January 31, 2015 to conduct a research related to project"Capacity building for post-graduate research on remote health monitoring inBangladesh". I worked with Prof Nalini Venkatasubramanium at DistributedMiddleware Services (DMS) lab of Information and Computer Sciences departmentof UCI. The visit was quite great and a set of interesting problems was studiedduring the visit.
1.1 University of California IrvineUC Irvine, a research school, is one of the ten campuses of University of California.
Among ten UC system campuses this campus is relatively new (establishedaround 1965) and is regarded one of the most prominent campuses among 50years olds universities in the United States. The 1474 acre campus located inOrange county California hosts nearly 24 thousand students both undergrads andpost-graduates. The campus has beautiful landscape and its various schools anddepartments organize themselves in a radial fashion around a park, named theAldrich Park. Figure 1: UC Irvine campus (Aldrich Park is at the center) 1.2 Information and Computer Sciences DepartmentThe department of Information and Computer Sciences is located in a modernstate-of-the-art 6-storied building named Donald Bren Hall (DBH) located in HenrySimueli School of Engineering campus. The building houses three departments,namely statistics, informatics and Information and Computer science. As of year2014, they have a total of 68 faculties, 40 staffs, 1168 undergrads and 450
graduate students. UCI also has an Electrical Engineering and Science department located in Engineering Hall.
2. Research conductedThe research conducted during the visit mainly lies inthe direction of constructing building blocks formiddleware services for media content delivery incrowdsourcing applications tailored for m-health.
Recently media social networks emerged that enableend users upload and share rich media content from their mobile phones in theform of images, audio, and video. A few of them lately emerged in m-healthcontext. Figure1 is an example. Figure1 provides a mobile appthat allows doctors and medical professionals to capture photos of particularmedical cases of patients currently under their treatment and share those photosonline to seek options, comments and suggestions from community members,especially from other doctors and experts. These images capture real cases withvisual detail and supporting texts seeking need non-trivial attention from healthprofessionals to comment on and to provide necessary suggestions.
Figure 2: Home page of "Figure1" These kind of services in general involve interaction in terms of engagementamong users and delivery of media content between back-end servers andclients, where the clients are usually hosted at resource constrained mobiledevices. Although the service modality requires the end users to have constantconnections to the Internet for receiving live feeds and delivering contents, theassumption may not hold for many users for obvious reasons. In this work, wepropose a middleware service to be placed in between the server and the clientsthat would handle content delivery even though the mobile device may remainoffline for unpredictable time. The underlying problems eventually boil tomodeling of a content queue at the middleware (with its associated queuedynamics), and a set of solutions are suggested in the form control algorithmsusing Lyapunov optimization framework. In that report I will briefly highlight keyelements of the investigation and the framework used. 2.1 Media Crowdsourcing Media crowdsourcing enables individuals to generate rich media content andshare them with others through a platform, called crowdsourcing platform. Mediacontents are high fidelity contents that have more expressive power than usualtext. Media contents are abundant these days. A few names can be Instagram: shares photos Soundcloud: stores and shares songs, supports music streaming.
Youtube, vimeo: contains videos.
Facebook: enables sharing photos, photo albums, videos, etc.
Most of these applications are examples of publish-subscribe systems wherecontents are published (by ordinary users) and are then pushed to clients whosubscribed to those contents. Clients are called subscribers. Our works extendmiddleware services for this kind of pub-sub systems.
One important aspect of media contents make them tricky to handle. That is theyare usually large in size. So it engages significant resources to fetch, process,store and disseminate media contents. To circumvent that users are usually"notified" of the media contents via notifications (e.g., push notifications inphones) about their availability. Upon receiving notifications, users can then goand invoke retrieval if they want to do so. Notifications can be made rich so thatthey can contain some part of content if suitable.
Media contents are usually published in one exact format and size (e.g., anuploaded photo). But it can be argued that it is not always required to deliver thesame exact content to all users for consumption. Contents might have differentpresentations. For example, photos can be presented in low resolution, inthumbnail format, text description or even as a caption. Photo albums can bereplaced with smaller number of representative photos. These changes to contentare referred to as "content adaptation". Although media content is good to have,processing them on mobile devices pose further challenges. Especially becausemobile devices are resource constrained. They have limitations in terms ofenergy, storage, bandwidth, and network connections. It is interesting to note thatbased on user context and available resources, media content can be "adapted"to different presentations (may be degraded in quality). This adaptation may bedirected in order to optimize resource consumption as well as to maximize user'ssatisfaction in consuming delivered contents 2.2 Problem Formulation The following problem is formulated based on content adaptation. Deliverrelevant media contents optimally to mobile users. The set of question we need toanswer are – which contents? at what level of details? when, now? or can be delayed?
utility-cost trade-offs… More particularly, we ask questions and try to come with techniques that can beused to answer them satisfactorily. Which contents are relevant? These may bedepended on matching preferences, interests, expertise, social ties, etc. Contentsneed to presented at what granularity/presentation level? Different presentationshave different utilities. For example, original content (no degradations) has thelargest utility and utility declines as content gets degraded. The next thing toconsider is the availability of resources (e.g., energy/bandwidth budget forfetching contents). The problem is then to ask---given a set of contents with theirutilities, find a subset of contents that maximize total utility achieved from theirdelivery subject to various resource constraints. 2.3 Middleware for Content Adaptation and DeliveryWe build a middleware to adapt contents as suggested and deliver them to endusers. Middleware is a sort of a staging area for contents before they are deliveredto users. In terms of operation, it selects most relevant contents per user and thenselects suitable presentations per contents that are selected. Figure 3: Middleware architecture 2.3.1 Model of the MiddlewareWe model the middleware as a collection of content queues. Contents arrive in thequeues as they are published and are fetched to deliver to the subscribers. Andcontents depart from the queues as they are delivered to the users (mobiledevices) or dropped. The goal of the operation is to maintain a "stable" queuewhile maximizing the sum of utility of delivered contents. By stability we meanqueue should not grow unbounded (i.e., the expected queue length should remainfinite).
As per classical queuing theory, a (M/M/1) queue remains stable if service rate (µ)is strictly higher than arrival rate (λ), that is µ > λ. The problem is we don't reallyknow these two rates. That is, the parameters for arrival process and departureprocess are unknown. Instead, the system can be modeled as running in a seriesof intervals. And for each interval, t, we have— A(t): new contents arrive during interval t µ(t): contents scheduled to depart during interval t Using these two parameters for each slot, we can formulate a queue stabilizingcontrol algorithm based on Lyapunov framework to optimize our requiredobjective of maximizing total utility.
2.3.2 Lyapunov FrameworkLyapunov framework (LF) is a "control strategy" that helps achieving queuestability amid of unknown queue dynamics (arrival and departure). Lyapunovoptimization framework enables to design control algorithm that optimizes certainobjective function subject to keeping average queue length finite. It guarantees toachieve near-optimal performance while keeping queue stable.
The core concept in Lyapunov stability is called Lyapunov drift. It refers tomeasuring a running quantity as the queue grows. Let Q(t) be the queue backlogat time t. We define a Lyapunov function as L(t) = Q2(t)/2 (division by 2 is madefor mathematical simplicity that follows). Lyapunov drift, denoted as ∆(t),measures the difference between the values of Lyapunov functions in twosuccessive time intervals. That is, ∆(t) = L(t+1) – L(t). Control strategy is to keepthis drift as small as possible. A queue is said to be "stable" it the long-run average of the queue backlogremains finite. That is— Q=lim ¿ ∑ Q(τ )<∞ The following theorem is the main results for queue stability.
Theorem (Lyapunov Stability). If there exists constants B, ϵ , such that for
all timeslots we have—
E [∆ (t ) Q(t)¿ ≤ B−ϵQ(t ) then, the queue is stable, that is ´ Q=lim ¿ ∑ Q(τ )<∞ , and furthermore, ´Q≤ The proof is surprisingly simple, which makes use of iterated expectation andtelescopic sum.
The control strategy then tries to minimize E [∆ (t ) Q(t)¿ subject to requiredconstraints. Since the values of A(t)'s are beyond the controller's hand, thestrategy chooses appropriate µ(t)'s to achieve stability. It would be information ifwe show how values of B and ϵ are deducted. The framework assumes that thesecond moment of both arrival and departure process remain bounded. Let λ be the upper envelope of A(t) and µ be the lower envelop of µ(t), that is, for all t, A(t)≤ λ and µ ≥ µ(t). It can be shown that B = λ 2 + µ2. Again, for arrival process λ, ifthere exists an optimal control strategy with departure rate µ' that stabilizes thequeue, then ϵ = µ' - λ.
Lyapunov framework also supports optimization of certain objective function whilekeeping queue stable. Let this objective be some "penalty" that LF minimizes.
One example of this penalty can be energy consumption that the system tries tominimize over time. In that, drift is defined as ∆(t) + V P(t), for a controlparameter V. So, control strategy now tries to minimize E[∆ (t )+VP(t)∨Q (t )] The parameter V makes a trade-off between P and Q. The following theoremestablishes the results.
Theorem (Lyapunov stability with penalty). If there exists B and ϵ such
that for all slots, we have—
E [∆ (t )+VP (t ) Q(t)¿ ≤ B− Q ϵ (t )+V P¿ then the queue remain stable and the followings hold. P≤ P¿+ V Q≤ +V P¿ It can be observed that average P can be made arbitrarily close to optimal P* byenlarging V but that lets the queue grow. While P decreases at the rate of O(1/V),queue grows at a rate of O(V). This is called [O(1/V), O(V)] trade-off in Lyapunovframework. As we can see, V is an important "knob" to tune performance of the scheme. In effect, V controls the size of the queue (storage overhead at themiddleware). V can be a constant or can be adjusted according to certainconditions (e.g., network connectivity). 3. ConclusionIn this work a middleware service for content adaptation for media contentdelivery to mobile devices is developed. An optimization formulation is madeusing Lyapunov framework and solved with heuristics. Experiments are currentlyunderway. We consider only subscription side of the service that handles reception of contents. Similar efforts can be given on publisher side for uploadingcontents.
4. AcknowledgementThis work is a joint effort with other researchers, namely Vinay Setty from MaxPlanck Institute, Germany, Ye Zhao from Google, Mountain View, CA, and ProfRoman Vitenberg of University of Oslo, Norway. The author conveys heartiestgratitude to them. He also expresses his sincere thanks to all members of DMSlab, namely Kyle, Charles, Gene, Andy, Quixi, and Ranga, for their supports duringhis stay at UC Irvine. And lastly not the least, special thanks to ProfVenkatasubramanium for being a generous host of the visit.
A.G. Reznikov, N.D. Nosenko, L.V. Tarasenko, P.V. Sinitsyn, A.A. Lymareva Prenatal dexamethasone prevents early and long-lasting neuroendocrine and behavioral effects of maternal stress on male offspring Â ðîáîò³ ïåðåâ³ðåíî ã³ïîòåçó ïðî ðîëü ãîðìîí³â ã³ïîòàëàìî-ã³ïîô³çàðíî-àäðåíîêîðòèêàëüíî¿