To capture transient dynamics underlying neural changes of task-switchingtraining, we provided participants with a modified version of the AX continuous performance task ( Braver et al., 2005 ; Lenartowicz et al., 2010 ), which was adapted from the study by Schmitt et al. (2014) using pictures instead of letters as stimuli (see also Figure 2A). In this paradigm, participants were presented in each trial with a face picture (cue stimulus: a young or old face, which was either male or female 2 ) that was followed by an animal picture (probe stimulus: bird, cat, fish, or rabbit 3 ), where the latter had to be responded to by a left or right button press. The cue-probe combinations were presented in two types of conditions: context-dependent (c-dep) and context- independent (c-indep) trials. On c-dep trial conditions, the correct answer to the probe stimulus was dependent on the preceding cue information. For instance, participants had to press the left button in response to the bird stimulus, and the right button in response to the cat stimulus, if the preceding cue picture had been a young woman. If the same probes, however, followed the cue picture of an old man, the response mappings had to be reversed. In contrast, on c-indep trial conditions, correct responses to the probe stimuli were independent of the preceding cue. Hence, subjects always had to press the same response button to a given probe stimulus (e.g., left button press for the fish picture), irrespective of the preceding cue (i.e., old- woman or young-man picture). Importantly, c-dep as compared to c-indep conditions (further referred to as ‘context-updating costs’) put high demands on the flexible updating of stimulus- response mappings from trial to trial, thus pointing to transient processing dynamics.
Theoretical background and design. In Paper IV, we provided evidence for spatio-temporal interactions of neural plasticity associated with improvements in the task-switchingtask. More specifically, we revealed selective changes in block-related brain activation in fronto-polar regions and in the basal ganglia, while in trial-related brain activation in fronto-lateral and parietal regions after task-switchingtraining. Based on these findings, in a next step, we try to determine whether such spatio-temporal interactions also modulate the neural transfer of training to structurally dissimilar cognitive tasks. In general, neural transfer has been assumed to occur if activation changes associated with the training and transfer tasks rely on the same cognitive processes and on spatially overlapping brain regions (Dahlin, Neely, Larsson, Bäckman, & Nyberg, 2009). However, it has not yet been examined whether the amount of transfer also depends on temporal overlap, that is, on overlapping sustained or transient timescales of the involved neural processes. Therefore, in the current project, we aim to dissolve interactions between spatial and temporal dynamics of brain mechanisms supporting transfer. In our study in older adults, we had also administered two transfer paradigms, either sensitive to capture sustained dynamics of task-switching behavior (i.e., a delayed-recognition working- memory task, adapted from Clapp, Rubens, & Gazzaley, 2010) or to capture transient dynamics (i.e., a context-updating task, adapted from Schmitt, Ferdinand, & Kray, 2014a). These were approached by the appropriate block- related or event-related fMRI design. Neural transfer is here defined as selective changes in task activation during the delayed-recognition or context-updating transfer task after task-switchingtraining compared to single-task control training. In respect of the Lövdén et al. (2010) framework, the same design principles as in Paper IV may apply here, except that we tested a broader scope of
group, that is, if participants were allowed to verbalize at posttest (and not only during training), then transfer may occur. In order to test this latter hypothesis, an additional experiment was performed (Karbach & Kray, in prep.). Thirty-eight older participants (mean age = 68.2 years of age) were investigated in a pretest – training – posttest design. At pretest and posttest, participants performed an internally cued switching paradigm including two tasks A and B that were similar to those applied in the present study 50 . In the training phase, all participants received task-switchingtraining (including tasks C and D, also similar to those applied in the present study) and verbal self-instruction training. Importantly, half of the participants were instructed to continue using the verbalization strategy at posttest, while the other half was not allowed to verbalize. This second group corresponds to the group receiving task-switching and verbal self-instruction training in the present study. Results from this so far unpublished study showed that participants which continued verbalizing at posttest showed a larger reduction of general switch costs (337 ms to 103 ms) and specific switch costs (347 ms to 189 ms) than the group that was not allowed to verbalize at posttest (general switch costs: 270 ms to 155 ms; specific switch costs: 251 ms to 181 ms). Thus, when both the training and transfer situation allow the application of the verbal strategy, verbal self-instruction benefits can be transferred to a new, similar switchingtask, at least in older adults. This finding is in line with results from Healy, Wohldmann, Parker, and Bourne (2005) showing that participants performing a secondary verbal task during training in a prospective paradigm performed worse during transfer when the secondary task was not required during transfer. The authors suggested that the trainingtask and the verbal task are integrated into a single, more complex task during practice and that transfer only occurs when the cognitive operations acquired during training can be used in the same way during transfer.
increased. There are theories suggesting that older adults have difficulties when task uncertainty increases due to lack of contextual cues from the environment (Lindenberger & Mayr, 2013). In agreement with previous studies (e.g., Kray et al., 2002) our results from the four training sessions showed larger switch costs in uncued groups than in cued groups, but no age differences therein. While some studies indicated that older adults can use task cues to prepare for an upcoming task in a similar way as younger adults as evidenced by a similar amount of reduction of switch costs with changing cue stimulus interval (Cepeda et al., 2001; Kray, 2006), there is also a study showing age-related impairments in mixing costs when time for task preparation is decreased (Lawo et al., 2012). Our results showed a similar amount of reduction in switch costs in younger and older adults throughout the four training sessions in both cued and uncued groups, which implies that switching ability might be independent of contextual demands. Furthermore, it has been demonstrated that in an alternate runs task-switching paradigm if updating demands are increased then older adults are less able to reduce switch costs through training (Kramer et al., 1999). Consistent with this finding we found larger mixing cost reductions in older adults in cued-bivalent task-switching groups than in uncued-bivalent task-switchingtraining groups.
The last question concerns the transferability of training effects to other, untrained cognitive domains or even to everyday life situations. Is there an additional performance improvement in tasks that were not part of the training intervention, but that call for the same cognitive mechanisms and therefore the same neural networks that have been trained? The issue of so-called “transfer effects” in the research literature is comprehensive, yet complex. Studies showed that the possibility of transfer effects often is limited in its extent and duration (Dahlin, Neely, Larsson, Bäckman, & Nyberg, 2008). There are various factors that can influence the probability of transfer, for example the structural similarity of the trained and untrained tasks (Rickard & Bourne, 1996). Transfer effects, in terms of improved cognitive performance in a transfer task, seem more likely to occur when the two tasks are similar (near transfer), than when they are dissimilar (Woodwarth & Thorndike, 1901). Nonetheless, far transfer is possible as long as the transfer task demands the trained cognitive functions (Gajewski & Falkenstein, 2012; Shipstead, Redick, & Engle, 2012). Literature on training and transfer effects of cognitive practice mainly focuses on processes of attention, memory, reasoning, or general cognitive control, showing that specific interventions were able to improve behavioral performance in older adults (e.g., Basak, Boot, Voss, & Kramer 2008; Bherer et al., 2005). However, there are less studies that use neuroimaging techniques like electroencephalography (EEG) or
Two specic experimental challenges have to be faced. Firstly, the applied pulses to the device need very short rise and fall times and must have a relatively high amplitude (>1 V), too. Commercial solutions that are able to generate such short pulses usually have amplitudes below 1 V and are only available as pattern generators with relatively high repetition rates. For probing the switching event, however, this is not favourable. Torrezan et al. solved this issue by an experi- mental setup that includes a commercial pattern generator, an additional amplier and a switch that separates one pulse from the generated pattern. 14 The fastest switching time of a ReRAM device has been measured with this setup and amounts to 85 ps. 19 The second challenge is to design the experiment in a way that the generated signal is preserved until it reaches the ReRAM device. Therefore, every component in the setup has to be chosen carefully and requires a high bandwidth. To provide proper impedance matching at the contact between the sample and the high frequency probes, coplanar waveguides are usually used, in which the ReRAM device is embedded (see Fig. 4). It consists of two ground conductors and the middle signal conductor. Its characteristic impedance can be designed by the dimensions of the line width and the spacing. Usually it is set to 50 U as this is the characteristic impedance of all other components in the setup. To provide ideal signal preservation, this transmission line needs to be termi- nated by 50 U in order to minimize eﬀects of parasitic capacitances and induc- tances. The resistance of a ReRAM device, however, is usually much higher and therefore those capacitive eﬀects still have to be considered. To illustrate this measurement technique, an exemplary measurement is provided in the following.
are presented in appendix A . In appendix B we show results when using different occupational classifications.
It is worth comparing the results as a whole against the logic of the bounds on task prices derived by Gottschalk
et al. ( 2016 ). They derive an upper bound on the increase in price on the abstract task compared to the routine manual task by computing statistics after trimming the top of the distribution of routine manual wages and the bottom of the abstract wage distribution. In their case they compute medians, but we compute means. Using the UK as an example, employment in the abstract occupation increased by around 13 percentage points, or a third of the 2008 total. Therefore we obtain a quick estimate of the upper bound by comparing the raw mean in the abstract occupation with the mean obtained from trimming a third of wages from the bottom. We do this for the abstract occupation in 2008 using residual wages after regressions on age and education to take account of observable factors. We find that the trimmed mean is 25 log points higher than the raw mean. We can compute an implied
Das Verfahren liefert neben einer Schätzung des Parame- tervektors θ zugleich eine Quantifizierung von Regime-Wahr- scheinlichkeiten in Abhängigkeit von der jeweils betrachte- ten Informationsmenge: Der Ausdruck p(s t = i | I T ) bezeich- net die bedingte Wahrscheinlichkeit, im Zeitpunkt t im Re- gime i zu sein, falls auf die gesamte Informationsmenge im Schätzzeitraum [1,…,T] des MS-Modells konditioniert wird (geglättete Wahrscheinlichkeit). 4 Der Ausdruck p(s t = i | I t ) gibt dagegen die bedingte Wahrscheinlichkeit für Zustand i an, falls nur auf die bis zur Rechenperiode t vorliegenden Informationsmenge fokussiert wird (gefilterte Wahrschein- lichkeit). Letztere ist unter Echtzeit-Aspekten besonders in- teressant. Für den Endzeitpunkt T stimmt der gefilterte Wert mit dem geglätteten überein. Die geglätteten Wahrschein- lichkeiten sind insbesondere dazu geeignet, die Dynamik der untersuchten Zeitreihen ex post zu untersuchen. Dadurch lassen sich in der Rückschau, bei Verwendung der gesam- ten Informationsmenge, Regimewechsel zuverlässig datie- ren. Denn bis auf den Rand der Zeitreihe, sind bei den Be- rechnungen zu allen Zeitpunkten sowohl die Vergangenheit als auch die Zukunft bekannt. Die Situation, in der sich der Konjunkturprognostiker befindet, wird dagegen in den ge- filterten Wahrscheinlichkeiten nachempfunden. Über den ge- samten Untersuchungszeitraum hinweg fließen in die Be- rechnung der Zustandswahrscheinlichkeiten jeweils nur die Daten aus der Vergangenheit ein. 5 Dies erhöht die Unsicher- heit bei der Bewertung, in welchem Regime sich der Pro- zess befindet. Gerade in dieser Situation sollen die Mar- kov-Switching-Modelle aber zusätzliche Entscheidungshil- fen geben.
“step in” at a point in which agents already hold some initially-signed contracts. An interesting avenue for future research is to generalize this and model also the interaction with the task providers. An intermediate step towards this may be to study a model in which agents may choose not to complete some tasks.
Thus far, we have not discussed who chooses the mechanism but rather argued in favor of a particular mechanism. Again, it is intrinsic to the open, blockchain-based economy that anyone can set up a smart contract. This leaves little room for rent extraction for the contract creator as anyone can copy the contract, make modifications, and invite the service providers to settle their allocation on the alternative platform. Thus, we expect that whether a mechanism will thrive in practice will depend on the fairness of the solution it provides. The reasoning is entirely analogous to what has been observed for centralized matching mechanisms, which, as noted for instance by Roth (1991), are most often successful when the outcomes they produce are perceived as fair (“stable”). Hence, similar to the evolution of norms in society, we expect to converge towards a mechanism that ensures cost-effective allocations with fair sharing of the cost reductions.
The rapid progress of electronics over the past decades has altered many aspects of our lives fundamentally. As the conventional concepts for non-volatile memory technologies are approaching their physical scaling limits, new memory concepts are demanded. One of the most promising approaches to push the scalability limit further is the redox-based resistive switching random access memory (ReRAM). The resistive switching in these ReRAM memory devices is controlled by applying appropriate voltage pulses, whereby the information is stored as the resistance level of the device. Its simple, two-terminal structure in combination with a strong local confinement of the device region responsible for resistive switching allow for a remarkable scaling potential of ReRAM technology. Together with a high potential of energy efficiency, fast access and low cost it is considered as a potential competitor for both Flash and DRAM. Resistive switching is observed in various material systems, however, many questions regarding interplay between different materials and its impact on switching properties remain unclear.
English. In this paper, we propose a clas- sifier for predicting sentiments of Italian Twitter messages. This work builds upon a deep learning approach where we lever- age large amounts of weakly labelled data to train a 2-layer convolutional neural net- work. To train our network we apply a form of multi-tasktraining. Our system participated in the EvalItalia-2016 com- petition and outperformed all other ap- proaches on the sentiment analysis task. In questo articolo, presentiamo un sis- tema per la classificazione di soggettivit`a e polarit`a di tweet in lingua italiana. L’approccio descritto si basa su reti neu- rali. In particolare, utilizziamo un dataset di 300M di tweet per addestrare una con- volutional neural network. Il sistema `e stato addestrato e valutato sui dati for- niti dagli organizzatori di Sentipolc, task di sentiment analysis su Twitter organiz- zato nell’ambito di Evalita 2016..
Farmer et al. (2009) point out that the posterior distribution might be highly non-Gaussian and the mean values of this distribution may actually lie in a region where the support is flat. Hence, it is of interest to search for the posterior mode rather than the mean. This task, however, may be computationally intensive, as the posterior is often multi-modal and the optimization algorithm may get stuck at a local mode. Farmer et al. (2009) propose a specific block-wise optimization algorithm to deal with the problem, while Sargent et al. (2009) use a Gibbs sampling version of Chris Sims’ CSMINWEL routine, followed by a combination of the BFGS Quasi-Newton algorithm and Fortran’s IMSL routine. Nevertheless, Liu and Mumtaz (2011) and Chen and Macdonald (2012) report successful usage of CSMINWEL alone. For the estimation at hand, Sims’ routine faced particular difficulties finding the global mode and often got stuck at local maxima with a high likelihood value (as the procedure is actually a minimization algorithm, high values are undesirable). For maximization of the likelihood function the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) has been employed, particularly its extension for DSGE models by Martin Andreasen [Andreasen (2008)]. The procedure is based on evolution strategy algorithms, which do not calculate gradients or approximate numerical derivatives. This is a considerable advantage when the target functions have discontinuities, ridges, or local optima. [Hansen (2006)].
used to obtain equation (equation (12)). The decreased inter- face exchange barrier originates from the Butler–Volmer model (equation (10)) as ΔW e + βeΔV act . For the cell with a direct SrTiO 3 interface to platinum, the slope of the voltage drop over the interfacial Schottky barrier decreases for voltages above 2 V, as the voltage increasingly drops over the SrTiO 3 series resist- ance. The ion migration and interface exchange barriers are not lowered further (black and red dashed lines in Figure 3a) and the switching kinetics flatten (compare Figure 2a). For the cell with an Al 2 O 3 interlayer, on the other hand, less voltage drops over the SrTiO 3 far away from the interface but instead over the interlayer (green solid line in Figure 3a). This leads to a steady decrease of the ion migration barrier in this interface layer (green dashed line) and goes together with the continuously decreasing switching times in the switching kinetics of this cell type (compare Figure 2c). The energy barrier for the interface exchange reaction in the cells with an Al 2 O 3 interlayer is much higher than the migration barriers in the SrTiO 3 and the Al 2 O 3 layer. But since it is sufficient that the oxygen is released from
[ 2012 ] who study the welfare implications of HBPD under asymmetric market shares. The authors find that even when firms can price discriminate between new and current customers, poaching might not take place if switching costs are sufficiently high. Moreover, where market shares are particularly skewed, the erosion of the larger firm’s customer base is larger than under uniform pricing. Indeed, for very asymmetric inherited market shares the larger firm may not offer a poaching offer given that it would be too costly to attract marginal customers that are close to the previous cut-off point, but very far from the opposite extreme where the dominant firm is located. This is because, similar to Thisse and Vives [ 1988 ], the authors also include brand preferences under a linear Hotelling model so that it is too expensive for the larger firm to pre-empt poaching of its least loyal customers.
There are some different proposed models/theories trying to explain the resistive switching in MIM devices based on electrical effects. Electronic charge injection and/or charge displacement is seen as one origin of the switching. As early as 1967, Simmons and Verderber proposed a charge–trap model which can explain the different resistance states of a MIM cell by the modification of the electrostatic barriers due to trapped charges . According to this model, charges are injected by Fowler–Nordheim tunneling at high electric fields and subsequently trapped at sites such as defects or metal nanoparticles in the insulator. This model was later modified in order to incorporate charge trapping at interface states, which is thought to affect the adjacent Schottky barrier at various metal/semiconducting perovskite interfaces . Another proposed explanation for perovskite–type oxides is based on the insulator–metal transition (IMT) . In this model, the electronic charge injection acts like doping, which induces an IMT [39, 40].
A qualitatively similar picture is found in the case of a rotating field with angular frequency directed opposite to the vortex polarity: wp < 0. There exist different regimes in the vortex dynamics . In the range of frequencies close to the frequency of the orbital vortex motion ( W ~1 GHz the vortex demonstrates a finite motion in a ) region near the disk center along a quite complicated tra- jectory and does not switch its polarity. These cycloidal vortex oscillations are similar to those in Ref. , and correspond to the excitation of higher magnon modes dur- ing the motion. If w >> W, then for weak fields the vortex motion can be considered as a sum of two constituents: (i) the gyroscopic orbital motion as without field, and (ii) cy- cloidal oscillations caused by the field influence. For the case wp < 0 the direction of the cycloidal oscillations coin- cides with the direction of the field rotation, while the di- rection of the gyroscopic orbital motion is opposite to it. For stronger fields the vortex motion becomes more com- plicated and its average motion can be directed even op- posite to the gyroscopical motion (this situation is shown in Fig. 1). The irreversible switching of the vortex polar- ity can be excited in a specific range of parameters ( , ) w b , with typical frequencies about 10 GHz and intensities about 20 mT . The mechanism of the switching is discussed below.
In its attempt to highlight the mechanism created by switching costs as a generator of a relationship between the intensity of competition and financial stability in banking markets, our model has abstracted from a number of important factors that should be incorporated in a future analysis. For example, we have characterized the return distributions associated the banks´ investments (lending activities) as an exogenous feature of our model. However, it is likely that changes in switching costs affect the banks’ lending and investment decisions in ways reminding of the effects of changes in various regulatory policies (see, e.g., Hellman et al. (2000) and Repullo (2004)) for the effects of capital requirements and Moreno and Takalo (2015) for the effects of transparency regulation on banks’ asset risk taking). Likewise, Matutes and Vives (2000) show how imperfect competition in the deposit market together with limited liability and deposit insurance affects the banks’ asset risk-taking incentives. Our model could be modified to analyse the effects of (foreign) entry in configurations where all depositors belong to the inherited market segment of the incumbent bank. Earlier studies (see, e.g., Sengupta (2007)) have highlighted the benefits of foreign entry in banking markets. However, if the domestic bank’s depositors face switching costs, increased competition via foreign entry would not only lead to higher deposit rates, but potentially also to less stable banking market, in line with the casual evidence from, for example, the effects of cross-border entry of Icelandic banks in various markets in the early years of the 2000s.
Reasons for the beneficial effect of task prioritization can be explained by limited cognitive capacities (i.e., “single channel model”; Pashler, 1994; Pashler and Johnston, 1998 ) and/or cognitive interference when two tasks share the same processing resources (i.e., “capacity sharing model”; Tombu and Jolicoeur, 2003; Wickens, 2008 ). Particularly, the latter theory is well- suited to explain positive effects of prioritization. The capacity sharing model argues that there is a pool of processing resources that can be distributed between different tasks. Whenever more processing resources are devoted to one task, limited processing capacity remains and tasks and performance deficits in the given tasks arise. By prioritizing one task over another, the limited resources are explicitly allocated and the prioritized task will benefit from this resource allocation. This model is also used by the “posture second” or the “posture first” ( Bloem et al., 2006; Yogev-Seligmann et al., 2012 ) strategy claiming that the cognitive task is prioritized over the motor task, withdrawing attention from controlling posture or that the motor task is prioritized over the cognitive task. During the latter, there is no limitation of motor performance and the risk of loosing balance and falling is low.
If consumers all had the same switching cost, they could make their pur- chasing decisions without taking into account the choices of other consumers. On the other hand, when switching costs differ, a firm’s future price will depend on the type of consumers to which it sold to in the past. Therefore, rational consumers take into account which consumers they expect to pur- chase from each firm when making their purchasing decision. This creates an externality across consumer types; as we will see, at equal prices, high switching cost consumers would prefer to purchase from a firm whose client` ele is mostly composed of low switching cost consumers.