Director and wife Suzy Amis also reveal that they went vegan after watching the documentary 'Forks Over Knives.'

From the Montepulciano region of Italy comes a rich, spicy wine from Italy’s largest winery, which is embracing biodynamic practices.

Method creates a healthy system of plants, animals and soil while considering celestial effects on life.

Boston Dynamics robots imitate human and animal movements, making them impressive — and a little creepy.

Scandinavian based company Jordi Hans Design will be launching a crowdfunding campaign for their cargo bike trailer called SideBuddy on Kickstarter this June.

Atlanta-based Nonprofit Hosts Enrollment Period for 2020 Cohort

New method of training AI poised to eliminate the need for expensive cloud infrastructure and bring the era of ubiquitous, human-like artificial intelligence a decade closer.

Location Eglinton Ave West and Dufferin Steer in Toronto. Our experienced professional staff is driven to help you achieve optimal recovery. We diagnose and treat a wide variety of injuries including sport injuries, injuries associated with workplace

New research on streaming data systems will identify trends in analytics and data management

New research on engaging candidates effectively will guide HR teams and business leaders in this critical arena

The Relationship is a Perfect Fit for Honest Tax's Business Philosophy of Providing Clients with Honest Answers and Honest Results

Item by Item (IBI) – www.itembyitem.co - announced today the release of Microsoft Dynamics 365 (D365) Commerce online training curriculum.

Latest research aims to understand advances in natural language capabilities and its impact on business

Customer data management solution cleans and enriches dynamics data and provides actionable insights

An analysis of energy efficiency and exergy effciency of a cogeneration system in a sugar refinery in São Paulo state in Brazil is presented in the ‘Análisis energético, exergético y económico de un sistema de cogeneración: caso para una planta azucarera de San Pablo’ by Omar R. Llerena of Universidade de São Paulo (Published in: Ingenius no.19 Jan/Feb 2018 under Creative Commons license CC BY-NC-SA 4.0)
[See image gallery at www.sankey-diagrams.com]
Flows are in kW, and even though this diagram appears to be ‘casero’ made from blocks, triangles and curve shapes, the flow widths seem pretty much to scale.
Acronym ‘CC’ is for the combustion chamber and ‘CR’ stands for a heat recovery boiler (caldera de recuperación).

The article also features a Grassman diagram for the exergy analysis. So, if you are interested, please visit the article here.

The Chugach and Tongass National Forests are changing, possibly in response to global warming.

A very simple method of magnifying an image using just CSS

We monitored coarse woody debris dynamics and natural tree regeneration over a 14-year period after the 1991 Warner Creek Fire, a 3631-ha (8,972-ac) mixed severity fire in the western Cascade Range of Oregon. Rates for tree mortality in the fire, postfire mortality, snag fall, and snag fragmentation all showed distinct patterns by tree diameter and species, with Douglas-fir (Pseudotsuga menziesii (Mirb.) Franco) more likely to survive a fire, and to remain standing as a snag, than other common tree species. Natural seedling regeneration was abundant, rapid, and highly variable in space. Densities of seedlings >10 cm height at 14 years postfire ranged from 1,530 to 392,000 per ha. Seedling establishment was not concentrated in a single year, and did not appear to be limited by the abundant growth of shrubs. The simultaneous processes of mortality, snag fall, and tree regeneration increased the variety of many measures of forest structure. The singular event of the fire has increased the structural diversity of the landscape.

We are pleased to announce a new Audio Inserts feature in the Auphonic API: audio inserts are separate audio files (like intros/outros), which will be inserted into your production at a defined offset.
This blog post shows how one can use this feature for Dynamic Ad Insertion and discusses other Audio Manipulation Methods of the Auphonic API.

API-only Feature

For the general podcasting hobbyist, or even for someone producing a regular podcast, the features that are accessible via our web interface are more than sufficient.

However, some of our users, like podcasting companies who integrate our services as part of their products, asked us for dynamic ad insertions. We teamed up with them to develop a way of making this work within the Auphonic API.

We are pleased therefore to announce audio inserts - a new feature that has been made part of our API. This feature is not available through the web interface though, it requires the use of our API.

Before we talk about audio inserts, let's talk about what you need to know about dynamic ad insertion!

Dynamic Ad Insertion

There are two ways of dealing with adverts within podcasts. In the first, adverts are recorded or edited into the podcast and are fixed, or baked in. The second method is to use dynamic insertion, whereby the adverts are not part of the podcast recording/file but can be inserted into the podcast afterwards, at any time.

This second approach would allow you to run new ad campaigns across your entire catalog of shows. As a podcaster this allows you to potentially generate new revenue from your old content.

As a hosting company, dynamic ad insertion allows you to choose up to date and relevant adverts across all the podcasts you host. You can make these adverts relevant by subject or location, for instance.

Your users can define the time for the ads and their podcast episode, you are then in control of the adverts you insert.

Audio Inserts in Auphonic

Whichever approach to adverts you are taking, using audio inserts can help you.

Audio inserts are separate audio files which will be inserted into your main single or multitrack production at your defined offset (in seconds).

When a separate audio file is inserted as part of your production, it creates a gap in the podcast audio file, shifting the audio back by the length of the insert. Helpfully, chapters and other time-based information like transcriptions are also shifted back when an insert is used.

The biggest advantage of this is that Auphonic will apply loudness normalization to the audio insert so, from an audio point of view, it matches the rest of the podcast.

Although created with dynamic ad insertion in mind, this feature can be used for any type of audio inserts: adverts, music songs, individual parts of a recording, etc. In the case of baked-in adverts, you could upload your already processed advert audio as an insert, without having to edit it into your podcast recording using a separate audio editing application.

Please note that audio inserts should already be edited and processed before using them in production. (This is usually the case with pre-recorded adverts anyway). The only algorithm that Auphonic applies to an audio insert is loudness normalization in order to match the loudness of the entire production. Auphonic does not add any other processing (i.e. no leveling, noise reduction etc).

Audio Inserts Coding Example

Here is a brief overview of how to use our API for audio inserts. Be warned, this section is coding heavy, so if this isn't your thing, feel free to move along to the next section!

You can add audio insert files with a call to https://auphonic.com/api/production/{uuid}/multi_input_files.json, where uuid is the UUID of your production.
Here is an example with two audio inserts from an https URL. The offset/position in the main audio file must be given in seconds:

curl -X POST -H "Content-Type: application/json" 
    https://auphonic.com/api/production/{uuid}/multi_input_files.json 
    -u username:password 
    -d '[
            {
                "input_file": "https://mydomain.com/my_audio_insert_1.wav",
                "type": "insert",
                "offset": 20.5
            },
            {
                "input_file": "https://mydomain.com/my_audio_insert_2.wav",
                "type": "insert",
                "offset": 120.3
            }
        ]'

More details showing how to use audio inserts in our API can be seen here.

Additional API Audio Manipulations

In addition to audio inserts, using the Auphonic API offers a number of other audio manipulation options, which are not available via the web interface:

Cut start/end of audio files: See Docs

In Single-track productions, this feature allows the user to cut the start and/or the end of the uploaded audio file. Crucially, time-based information such as chapters etc. will be shifted accordingly.

Fade In/Out time of audio files: See Docs

This allows you to set the fade in/out time (in ms) at the start/end of output files. The default fade time is 100ms, but values can be set between 0ms and 5000ms.
This feature is also available in our Auphonic Leveler Desktop App.

Adding intro and outro: See Docs

Automatically add intros and outros to your main audio input file, as it is also available in our web interface.

Add multiple intros or outros: See Docs

Using our API, you can also add multiple intros or outros to a production. These intros or outros are played in series.

Overlapping intros/outros: See Docs

This feature allows intros/outros to overlap either the main audio or the following/previous intros/outros.

Conclusion

If you haven't explored our API already, the new audio inserts feature allows for greater flexibility and also dynamic ad insertion.
If you offer online services to podcasters, the Auphonic API would also then allow you to pass on Auphonic's audio processing algorithms to your customers.

If this is of interest to you or you have any new feature suggestions that you feel could benefit your company, please get in touch. We are always happy to extend the functionality of our products!

If listeners find themselves using the volume up and down buttons a lot, level differences within your podcast or audio file are too big.
In this article, we are discussing why audio dynamic range processing (or leveling) is more important than loudness normalization, why it depends on factors like the listening environment and the individual character of the content, and why the loudness range descriptor (LRA) is only reliable for speech programs.

Photo by Alexey Ruban.

Why loudness normalization is not enough

Everybody who has lived in an apartment building knows the problem: you want to enjoy a movie late at night, but you're constantly on the edge - not only because of the thrilling story, but because your index finger is hovering over the volume down button of your remote. The next loud sound effect is going to come sooner rather than later, and you want to avoid waking up your neighbors with some gunshot sounds blasting from your TV.

In our previous post, we talked about the overall loudness of a production. While that's certainly important to keep in mind, the loudness target is only an average value, ignoring how much the loudness varies within a production. The loudness target of your movie might be in the ideal range, yet the level differences between a gunshot and someone whispering can still be enormous - having you turn the volume down for the former and up for the latter.

While the average loudness might be perfect, level differences can lead to an unpleasant listening experience.

Of course, this doesn't apply to movies alone. The image above shows a podcast or radio production. The loud section is music, the very quiet section just breathing, and the remaining sections are different voices.

To be clear, we're not saying that the above example is problematic per se. There are many situations, where a big difference in levels - a high dynamic range - is justified: for instance, in a movie theater, optimized for listening and without any outside noise, or in classical music.
Also, if the dynamic range is too small, listening can be tiring.

But if you watch the same movie in an outdoor screening in the summer on a beach next to the crashing waves or in the middle of a noisy city, it can be tricky to hear the softer parts.
Spoken word usually has a smaller dynamic range, and if you produce your podcast for a target audience of train or car commuters, the dynamic range should be even smaller, adjusting for the listening situation.

Therefore, hitting the loudness target has less impact on the listening experience than level differences (dynamic range) within one file!
What makes a suitable dynamic range does not only depend on the listening environment, but also on the nature of the content itself. If the dynamic range is too small, the audio can be tiring to listen to, whereas more variability in levels can make a program more interesting, but might not work in all environments, such as a noisy car.

Dynamic range experiment in a car

Wolfgang Rein, audio technician at SWR, a public broadcaster in Germany, did an experiment to test how drivers react to programs with different dynamic ranges. They monitored to what level drivers set the car stereo depending on speed (thus noise level) and audio dynamic range.
While the results are preliminary, it seems like drivers set the volume as low as possible so that they can still understand the content, but don't get distracted by loud sounds.

As drivers adjust the volume to the loudest voice in a program, they won't understand quieter speakers in content with a high dynamic range anymore. To some degree and for short periods of time, they can compensate by focusing more on the radio program, but over time that's tiring. Therefore, if the loudness varies too much, drivers tend to switch to another program rather than adjusting the volume.
Similar results have been found in a study conducted by NPR Labs and Towson University.

On the other hand, the perception was different in pure music programs. When drivers set the volume according to louder parts, they weren't able to hear softer segments or the beginning of a song very well. But that did not matter to them as much and didn't make them want to turn up the volume or switch the program.

Listener's reaction in response to frequent loudness changes. (from John Kean, Eli Johnson, Dr. Ellyn Sheffield: Study of Audio Loudness Range for Consumers in Various Listening Modes and Ambient Noise Levels)

Loudness comfort zone

The reaction of drivers to variable loudness hints at something that BBC sound engineer Mike Thornton calls the loudness comfort zone.

Tests (...) have shown that if the short-term loudness stays within the "comfort zone" then the consumer doesn’t feel the need to reach for the remote control to adjust the volume.

In a blog post, he highlights how the series Blue Planet 2 and Planet Earth 2 might not always have been the easiest to listen to. The graph below shows an excerpt with very loud music, followed by commentary just at the bottom of the green comfort zone. Thornton writes: "with the volume set at a level that was comfortable when the music was playing we couldn’t always hear the excellent commentary from Sir David Attenborough and had to resort to turning on the subtitles to be sure we knew what Sir David was saying!"

Planet Earth 2 Loudness Plot Excerpt. Colored green: comfort zone of +3 to -5LU around the loudness target. (from Mike Thornton: BBC Blue Planet 2 Latest Show In Firing Line For Sound Issues - Are They Right?)

As already mentioned above, a good mix considers the maximum and minimum possible loudness in the target listening environment.
In a movie theater the loudness comfort zone is big (loudness can vary a lot), and loud music is part of the fun, while quiet scenes work just as well. The opposite was true in the aforementioned experiment with drivers, where the loudness comfort zone is much smaller and quiet voices are difficult to understand.

Hence, the loudness comfort zone determines how much dynamic range an audio signal can use in a specific listening environment.

How to measure dynamic range: LRA

When producing audio for various environments, it would be great to have a target value for dynamic range, (the difference between the smallest and largest signal values of an audio signal) as well. Then you could just set a dynamic range target, similarly to a loudness target.

Theoretically, the maximum possible dynamic range of a production is defined by the bit-depth of the audio format. A 16-bit recording can have a dynamic range of 96 dB; for 24-bit, it's 144 dB - which is well above the approx. 120 dB the human ear can handle. However, most of those bits are typically being used to get to a reasonable base volume. Picture a glass of water: you want it to be almost full, with some headroom so that it doesn't spill when there's a sudden movement, i.e. a bigger amplitude wave at the top.

Determining the dynamic range of a production is easier said than done, though. It depends on which signals are included in the measurement: for example, if something like background music or breathing should be considered at all.
The currently preferred method for broadcasting is called Loudness Range, LRA. It is measured in Loudness Units (LU), and takes into account everything between the 10th and the 95th percentile of a loudness distribution, after an additional gating method. In other words, the loudest 5% and quietest 10% of the audio signal are being ignored. This way, quiet breathing or an occasional loud sound effect won't affect the measurement.

Loudness distribution and LRA for the film 'The Matrix'. Figure from EBU Tech Doc 3343 (p.13).

However, the main difficulty is which signals should be included in the loudness range measurement and which ones should be gated. This is unfortunately often very subjective and difficult to define with a purely statistical method like LRA.

Where LRA falls short

Therefore, only pure speech programs give reliable LRA values that are comparable!
For instance, a typical LRA for news programs is 3 LU; for talks and discussions 5 LU is common. LRA values for features, radio dramas, movies or music very much depend on the individual character and might be in the range between 5 and 25 LU.

To further illustrate this, here are some typical LRA values, according to a paper by Thomas Lund (table 2):

Program	Loudness Range
Matrix, full movie	25.0
NBC Interstitials, Jan. 2008, all together (3:30)	9.4
Friends Episode 16	6.6
Speak Ref., Male, German, SQUAM Trk 54	6.2
Speak Ref., Female, French, SQUAM Trk 51	4.8
Speak Ref., Male, English, Sound Check	3.3
Wish You Were Here, Pink Floyd	22.1
Gilgamesh, Battle of Titans, Osaka Symph.	19.7
Don’t Cry For Me Arg., Sinead O’Conner	13.7
Beethoven Son in F, Op17, Kliegel & Tichman	12.0
Rock’n Roll Train, AC/DC	6.0
I.G.Y., Donald Fagen	3.6

LRA values of music are very unpredictable as well.
For instance, Tom Frampton measured the LRA of songs in multiple genres, and the differences within each genre are quite big. The ten pop songs that he analyzed varied in LRA between 3.7 and 12 LU, country songs between 3.6 and 14.9 LU. In the Electronic genre the individual LRAs were between 3.7 and 15.2 LU. Please see the tables at the bottom of his blog post for more details.

We at Auphonic also tried to base our Adaptive Leveler parameters on the LRA descriptor. Although it worked, it turned out that it is very difficult to set a loudness range target for diverse audio content, which does include speech, background sounds, music parts, etc. The results were not predictable and it was hard to find good target values. Therefore we developed our own algorithm to measure the dynamic range of audio signals.

In conclusion, LRA comparisons are only useful for productions with spoken word only and the LRA value is therefore not applicable as a general dynamic range target value. The more complex a production gets, the more difficult it is to make any judgment based on the LRA.
This is, because the definition of LRA is purely statistical. There's no smart measurement using classifiers that distinguish between music, speech, quiet breathing, background noises and other types of audio. One would need a more intelligent algorithm (as we use in our Adaptive Leveler), that knows which audio segments should be included and excluded from the measurement.

From theory to application: tools

Loudness and dynamic range clearly is a complicated topic. Luckily, there are tools that can help. To keep short-term loudness in range, a compressor can help control sudden changes in loudness - such as p-pops or consonants like t or k. To achieve a good mid-term loudness, i.e. a signal that doesn't go outside the comfort zone too much, a leveler is a good option. Or, just use a fader or manually adjust volume curves. And to make sure that separate productions sound consistent, loudness normalization is the way to go. We have covered all of this in-depth before.

Looking at the audio from above again, with an adaptive leveler applied it looks like this:

Leveler example. Output at the top, input with leveler envelope at the bottom.

Now, the voices are evened out and the music is at a comfortable level, while the breathing has not been touched at all.
We recently extended Auphonic's adaptive leveler, so that it is now possible to customize the dynamic range - please see adaptive leveler customization and advanced multitrack audio algorithms.
If you wanted to increase the loudness comfort zone (or dynamic range) of the standard preset by 10 dB (or LU), for example, the envelope would look like this:

Leveler with higher dynamic range, only touching sections with extremely low or extremely high loudness to fit into a specific loudness comfort zone.

When a production is done, our adaptive leveler uses classifiers to also calculate the integrated loudness and loudness range of dialog and music sections separately. This way it is possible to just compare the dialog LRA and loudness of complex productions.

Assessing the LRA and loudness of dialog and music separately.

Conclusion

Getting audio dynamics right is not easy. Yet, it is an important thing to keep in mind, because focusing on loudness normalization alone is not enough. In fact, hitting the loudness target often has less impact on the listening experience than level differences, i.e. audio dynamics.

If the dynamic range is too small, the audio can be tiring to listen to, whereas a bigger dynamic range can make a program more interesting, but might not work in loud environments, such as a noisy train.
Therefore, a good mix adapts the audio dynamic range according to the target listening environment (different loudness comfort zones in cinema, at home, in a car) and according to the nature of the content (radio feature, movie, podcast, music, etc.).

Furthermore, because the definition of the loudness range / LRA is purely statistical, only speech programs give reliable LRA values that are comparable.
More "intelligent" algorithms are in development, which use classifiers to decide which signals should be included and excluded from the dynamic range measurement.

If you understand German, take a look at our presentation about audio dynamic processing in podcasts for further information:

A topological dynamical system $(X,f)$ induces two natural systems, one is on the probability measure spaces and other one is on the hyperspace.

We introduce a concept for these two spaces, which is called entropy order, and prove that it coincides with topological entropy of $(X,f)$. We also consider the entropy order of an invariant measure and a variational principle is established.

The characterization of phase dynamics in coupled oscillators offers insights into fundamental phenomena in complex systems. To describe the collective dynamics in the oscillatory system, order parameters are often used but are insufficient for identifying more specific behaviors. We therefore propose a topological approach that constructs quantitative features describing the phase evolution of oscillators. Here, the phase data are mapped into a high-dimensional space at each time point, and topological features describing the shape of the data are subsequently extracted from the mapped points. We extend these features to time-variant topological features by considering the evolution time, which serves as an additional dimension in the topological-feature space. The resulting time-variant features provide crucial insights into the time evolution of phase dynamics. We combine these features with the machine learning kernel method to characterize the multicluster synchronized dynamics at a very early stage of the evolution. Furthermore, we demonstrate the usefulness of our method for qualitatively explaining chimera states, which are states of stably coexisting coherent and incoherent groups in systems of identical phase oscillators. The experimental results show that our method is generally better than those using order parameters, especially if only data on the early-stage dynamics are available.

The main aim of this paper is to detect dynamical properties of the Gy"orgyi-Field model of the Belousov-Zhabotinsky chemical reaction. The corresponding three-variable model given as a set of nonlinear ordinary differential equations depends on one parameter, the flow rate. As certain values of this parameter can give rise to chaos, the analysis was performed in order to identify different dynamics regimes. Dynamical properties were qualified and quantified using classical and also new techniques. Namely, phase portraits, bifurcation diagrams, the Fourier spectra analysis, the 0-1 test for chaos, and approximate entropy. The correlation between approximate entropy and the 0-1 test for chaos was observed and described in detail. Moreover, the three-stage system of nested subintervals of flow rates, for which in every level the 0-1 test for chaos and approximate entropy was computed, is showing the same pattern. The study leads to an open problem whether the set of flow rate parameters has Cantor like structure.

We study the time-averaged flux in a model of particles that randomly hop on a finite directed graph. In the limit as the number of particles and the time window go to infinity but the graph remains finite, the large-deviation rate functional of the average flux is given by a variational formulation involving paths of the density and flux. We give sufficient conditions under which the large deviations of a given time averaged flux is determined by paths that are constant in time. We then consider a class of models on a discrete ring for which it is possible to show that a better strategy is obtained producing a time-dependent path. This phenomenon, called a dynamical phase transition, is known to occur for some particle systems in the hydrodynamic scaling limit, which is thus extended to the setting of a finite graph.

We investigate the evolution in time of the position of a fixed number inthe insertion tableau when the Robinson-Schensted-Knuth algorithm is applied to asequence of random numbers. When the length of the sequence tends to infinity, a typical trajectory after scaling converges uniformly in probability to some deterministiccurve.

We live and cooperate in networks. However, links in networks only allow for pairwise interactions, thus making the framework suitable for dyadic games, but not for games that are played in groups of more than two players. To remedy this, we introduce higher-order interactions, where a link can connect more than two individuals, and study their evolutionary dynamics. We first consider a public goods game on a uniform hypergraph, showing that it corresponds to the replicator dynamics in the well-mixed limit, and providing an exact theoretical foundation to study cooperation in networked groups. We also extend the analysis to heterogeneous hypergraphs that describe interactions of groups of different sizes and characterize the evolution of cooperation in such cases. Finally, we apply our new formulation to study the nature of group dynamics in real systems, showing how to extract the actual dependence of the synergy factor on the size of a group from real-world collaboration data in science and technology. Our work is a first step towards the implementation of new actions to boost cooperation in social groups.

Uncertain dynamic obstacles, such as pedestrians or vehicles, pose a major challenge for optimal robot navigation with safety guarantees. Previous work on motion planning has followed two main strategies to provide a safe bound on an obstacle's space: a polyhedron, such as a cuboid, or a nonlinear differentiable surface, such as an ellipsoid. The former approach relies on disjunctive programming, which has a relatively high computational cost that grows exponentially with the number of obstacles. The latter approach needs to be linearized locally to find a tractable evaluation of the chance constraints, which dramatically reduces the remaining free space and leads to over-conservative trajectories or even unfeasibility. In this work, we present a hybrid approach that eludes the pitfalls of both strategies while maintaining the original safety guarantees. The key idea consists in obtaining a safe differentiable approximation for the disjunctive chance constraints bounding the obstacles. The resulting nonlinear optimization problem is free of chance constraint linearization and disjunctive programming, and therefore, it can be efficiently solved to meet fast real-time requirements with multiple obstacles. We validate our approach through mathematical proof, simulation and real experiments with an aerial robot using nonlinear model predictive control to avoid pedestrians.

A recently introduced classifier, called SS3, has shown to be well suited to deal with early risk detection (ERD) problems on text streams. It obtained state-of-the-art performance on early depression and anorexia detection on Reddit in the CLEF's eRisk open tasks. SS3 was created to deal with ERD problems naturally since: it supports incremental training and classification over text streams, and it can visually explain its rationale. However, SS3 processes the input using a bag-of-word model lacking the ability to recognize important word sequences. This aspect could negatively affect the classification performance and also reduces the descriptiveness of visual explanations. In the standard document classification field, it is very common to use word n-grams to try to overcome some of these limitations. Unfortunately, when working with text streams, using n-grams is not trivial since the system must learn and recognize which n-grams are important "on the fly". This paper introduces t-SS3, an extension of SS3 that allows it to recognize useful patterns over text streams dynamically. We evaluated our model in the eRisk 2017 and 2018 tasks on early depression and anorexia detection. Experimental results suggest that t-SS3 is able to improve both current results and the richness of visual explanations.

For real-time semantic video segmentation, most recent works utilised a dynamic framework with a key scheduler to make online key/non-key decisions. Some works used a fixed key scheduling policy, while others proposed adaptive key scheduling methods based on heuristic strategies, both of which may lead to suboptimal global performance. To overcome this limitation, we model the online key decision process in dynamic video segmentation as a deep reinforcement learning problem and learn an efficient and effective scheduling policy from expert information about decision history and from the process of maximising global return. Moreover, we study the application of dynamic video segmentation on face videos, a field that has not been investigated before. By evaluating on the 300VW dataset, we show that the performance of our reinforcement key scheduler outperforms that of various baselines in terms of both effective key selections and running speed. Further results on the Cityscapes dataset demonstrate that our proposed method can also generalise to other scenarios. To the best of our knowledge, this is the first work to use reinforcement learning for online key-frame decision in dynamic video segmentation, and also the first work on its application on face videos.

Restless multi-armed bandits are a class of discrete-time stochastic control problems which involve sequential decision making with a finite set of actions (set of arms). This paper studies a class of constrained restless multi-armed bandits (CRMAB). The constraints are in the form of time varying set of actions (set of available arms). This variation can be either stochastic or semi-deterministic. Given a set of arms, a fixed number of them can be chosen to be played in each decision interval. The play of each arm yields a state dependent reward. The current states of arms are partially observable through binary feedback signals from arms that are played. The current availability of arms is fully observable. The objective is to maximize long term cumulative reward. The uncertainty about future availability of arms along with partial state information makes this objective challenging. Applications for CRMAB abound in the domain of cyber-physical systems. This optimization problem is analyzed using Whittle's index policy. To this end, a constrained restless single-armed bandit is studied. It is shown to admit a threshold-type optimal policy, and is also indexable. An algorithm to compute Whittle's index is presented. Further, upper bounds on the value function are derived in order to estimate the degree of sub-optimality of various solutions. The simulation study compares the performance of Whittle's index, modified Whittle's index and myopic policies.

The accurate assembly of the system matrix is an important step in any code that solves partial differential equations on a mesh. We either explicitly set up a matrix, or we work in a matrix-free environment where we have to be able to quickly return matrix entries upon demand. Either way, the construction can become costly due to non-trivial material parameters entering the equations, multigrid codes requiring cascades of matrices that depend upon each other, or dynamic adaptive mesh refinement that necessitates the recomputation of matrix entries or the whole equation system throughout the solve. We propose that these constructions can be performed concurrently with the multigrid cycles. Initial geometric matrices and low accuracy integrations kickstart the multigrid, while improved assembly data is fed to the solver as and when it becomes available. The time to solution is improved as we eliminate an expensive preparation phase traditionally delaying the actual computation. We eliminate algorithmic latency. Furthermore, we desynchronise the assembly from the solution process. This anarchic increase of the concurrency level improves the scalability. Assembly routines are notoriously memory- and bandwidth-demanding. As we work with iteratively improving operator accuracies, we finally propose the use of a hierarchical, lossy compression scheme such that the memory footprint is brought down aggressively where the system matrix entries carry little information or are not yet available with high accuracy.

This paper proposes a faceted information exploration model that supports coarse-grained and fine-grained focusing of geographic maps by offering a graphical representation of data attributes within interactive widgets. The proposed approach enables (i) a multi-category projection of long-lasting geographic maps, based on the proposal of efficient facets for data exploration in sparse and noisy datasets, and (ii) an interactive representation of the search context based on widgets that support data visualization, faceted exploration, category-based information hiding and transparency of results at the same time. The integration of our model with a semantic representation of geographical knowledge supports the exploration of information retrieved from heterogeneous data sources, such as Public Open Data and OpenStreetMap. We evaluated our model with users in the OnToMap collaborative Web GIS. The experimental results show that, when working on geographic maps populated with multiple data categories, it outperforms simple category-based map projection and traditional faceted search tools, such as checkboxes, in both user performance and experience.

A proof of convergence is given for bulk--surface finite element semi-discretisation of the Cahn--Hilliard equation with Cahn--Hilliard-type dynamic boundary conditions in a smooth domain. The semi-discretisation is studied in the weak formulation as a second order system. Optimal-order uniform-in-time error estimates are shown in the $L^2$ and $H^1$ norms. The error estimates are based on a consistency and stability analysis. The proof of stability is performed in an abstract framework, based on energy estimates exploiting the anti-symmetric structure of the second order system. Numerical experiments illustrate the theoretical results.

Learning-based techniques are increasingly effective at controlling complex systems using data-driven models. However, most work done so far has focused on learning individual tasks or control laws. Hence, it is still a largely unaddressed research question how multiple tasks can be learned efficiently and simultaneously on the same system. In particular, no efficient state space exploration schemes have been designed for multi-task control settings. Using this research gap as our main motivation, we present an algorithm that approximates the smallest data set that needs to be collected in order to achieve high control performance for multiple learning-based control laws. We describe system uncertainty using a probabilistic Gaussian process model, which allows us to quantify the impact of potentially collected data on each learning-based controller. We then determine the optimal measurement locations by solving a stochastic optimization problem approximately. We show that, under reasonable assumptions, the approximate solution converges towards that of the exact problem. Additionally, we provide a numerical illustration of the proposed algorithm.

We provide a dynamical perspective on the classical problem of 3D point cloud registration with correspondences. A point cloud is considered as a rigid body consisting of particles. The problem of registering two point clouds is formulated as a dynamical system, where the dynamic model point cloud translates and rotates in a viscous environment towards the static scene point cloud, under forces and torques induced by virtual springs placed between each pair of corresponding points. We first show that the potential energy of the system recovers the objective function of the maximum likelihood estimation. We then adopt Lyapunov analysis, particularly the invariant set theorem, to analyze the rigid body dynamics and show that the system globally asymptotically tends towards the set of equilibrium points, where the globally optimal registration solution lies in. We conjecture that, besides the globally optimal equilibrium point, the system has either three or infinite "spurious" equilibrium points, and these spurious equilibria are all locally unstable. The case of three spurious equilibria corresponds to generic shape of the point cloud, while the case of infinite spurious equilibria happens when the point cloud exhibits symmetry. Therefore, simulating the dynamics with random perturbations guarantees to obtain the globally optimal registration solution. Numerical experiments support our analysis and conjecture.

Methods, apparatuses and articles of manufacture for use in a mobile device to determine whether a dynamic user profile is to transition from a first state to a second state based, at least in part, on one or more sensed indicators. The dynamic user profile may be indicative of one or more current inferable user behavior contexts for a user co-located with the mobile device. The mobile device may transition a dynamic user profile from a first state to a second state, in response to a determination that the dynamic user profile is to transition from the first state to the second state, and operatively affect one or more functions performed, at least in part, by the mobile device based, at least in part, on the transition of the dynamic user profile to the second state.

A system and method for reordering storage levels in a virtualized environment includes identifying a virtual machine (VM) to be transitioned and determining a new storage level order for the VM. The new storage level order reduces a VM live state during a transition, and accounts for hierarchical shared storage memory and criteria imposed by an application to reduce recovery operations after dynamic resource allocation actions. The new storage level order recommendation is propagated to VMs. The new storage level order applied in the VMs. A different storage-level order is recommended after the transition.

By employing a two-tier load balancing scheme, embodiments of the present invention may reduce the overhead of shared resource management, while increasing the potential aggregate throughput of a thread pool. As a result, the techniques presented herein may lead to increased performance in many computing environments, such as graphics intensive gaming.

In an example embodiment, there is described herein an apparatus comprising an interface for communicating with a plurality of digital signal processors and logic operable to send and receive data via the interface. The logic is configured to determine a first set of digital signal processors to be maintained in a ready state, a second set of digital signal processors to be maintained in a first energy saving state, and a third set of digital signal processors to be maintained in a second energy saving state.

A method for accurately determining the shape of currents in a current spectrum for a circuit design is provided. The method includes determining timing and power consumption characteristics. In one embodiment, timing characteristics are provided through a electronic design automation tool. The timing characteristics yield a current pulse time width. In another embodiment, power consumption characteristics are provided by an EDA tool. The power consumption characteristics yield a current pulse amplitude. The shape of the current pulse is obtained by incrementally processing a power analyzer tool over relatively small time increments over one or more clock cycles while capturing the switching nodes of a simulation of the circuit design for each time increment. In one embodiment, the time increments are one nanosecond or less.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for speech recognition. One of the methods includes receiving a base language model for speech recognition including a first word sequence having a base probability value; receiving a voice search query associated with a query context; determining that a customized language model is to be used when the query context satisfies one or more criteria associated with the customized language model; obtaining the customized language model, the customized language model including the first word sequence having an adjusted probability value being the base probability value adjusted according to the query context; and converting the voice search query to a text search query based on one or more probabilities, each of the probabilities corresponding to a word sequence in a group of one or more word sequences, the group including the first word sequence having the adjusted probability value.

A user interface help dialog for a computer system is provided in which the computer system has a plurality of system parameters and the user interface includes a plurality of command options that may be enabled or disabled. A cursor for indicating one of the command options is also provided. To generate the help dialog the identity of a disabled command option to which the cursor indicates is first identified. The status of a subset of the system parameters is then evaluated, the system parameters subset being determined in dependence on the identified command option. A help dialog to be displayed to the user is then generated, the content of which is dependent on the status of the evaluated system parameters.

There is provided a system and method for dynamically generated client side streaming playlists. There is provided a method comprising receiving a request to stream a video asset for playback, retrieving, from a network, a video asset playlist corresponding to the video asset and an ad campaign playlist, generating a consolidated video playlist including the video asset playlist and the ad campaign playlist, and processing the consolidated video playlist using a media playback framework to stream a plurality of video files from the network for decoding and output to a display. Discontinuity tags within the video asset playlist may be utilized as insertion points for portions of the ad campaign playlist. The consolidated video playlist may be accessed through a local HTTP web server, advantageously allowing the use of commonly installed media playback framework models such as QuickTime X to trigger adaptive bitrate support through HTTP Live Streaming.

Internet streaming from broadcast radio or television stations is described wherein triggers for dynamic content from internal or external systems cause an encoder system to generate command messages, and to optionally synchronize those command messages with any delays associated with the triggering events. Command messages are delivered through a streaming media distribution system to client media players which obtain or present the dynamic content, in association with any desired configuration changes to the appearance of the media player or the method or manner in which the dynamic content is presented.

Systems and methods for dynamic development and deployment of computing applications including a development framework, a visual design subsystem, and a deployment subsystem, where at runtime the deployment subsystem is operable to dynamically deploy a computing application realized by a blueprint by sending a request at runtime for graphs and components instantiated by the blueprint.

Systems and methods reconfigure data transmitted on a network, the network using a communication protocol. The systems and methods include a processor that is operable to receive data from a second device on the network; a memory medium; and a data configurator operable to dynamically reconfigure the data from the second device, with the data from the second device being in the form of a network item defined by the second device. The data configurator is operable to pass the reconfigured data to the processor and/or the memory medium in the form of a device item that is native to the device. A data design tool generates a map configuration to define a unique relationship between the network item and the device item. The map configuration includes a network representation portion for data in the network item and a device representation portion for data in the data item.

In a computer system that includes multiple nodes and multiple logical partitions, a dynamic partition manager computes current memory affinity and potential memory affinity to help determine whether a reallocation of resources between nodes may improve memory affinity for a logical partition or for the computer system. If so, the reallocation of resources is performed so memory affinity for the logical partition or computer system is improved. Memory affinity is computed relative to the physical layout of the resources according to a hardware domain hierarchy that includes a plurality of primary domains and a plurality of secondary domains.

A dynamically expandable and contractible fault-tolerant storage system employs a virtual hot spare that is created from unused storage capacity across a plurality of storage devices. This unused storage capacity is available if and when a storage device fails for storage of data recovered from the remaining storage device(s). On an ongoing basis, the storage system may determine the amount of unused storage capacity that would be required for the virtual hot spare (e.g., based on the number of storage devices, the capacities of the various storage devices, the amount of data stored, and the manner in which the data is stored) and generate a signal if additional storage capacity is needed for a virtual hot spare.

A method and apparatus are disclosed of monitoring a number of virtual machines operating in an enterprise network. One example method of operation may include identifying a number of virtual machines currently operating in an enterprise network and determining performance metrics for each of the virtual machines. The method may also include identifying at least one candidate virtual machine from the virtual machines to optimize its active application load and modifying the candidate virtual machine to change its active application load.

A method and system for dynamic distributed data caching is presented. The system includes one or more peer members and a master member. The master member and the one or more peer members form cache community for data storage. The master member is operable to select one of the one or more peer members to become a new master member. The master member is operable to update a peer list for the cache community by removing itself from the peer list. The master member is operable to send a nominate master message and an updated peer list to a peer member selected by the master member to become the new master member.