Finding good measures of robustness – i.e. the ability to correctly classify corrupted input signals – of a trained classifier is an important question for sensitive practical applications. In this paper, we point out that averaging the radius of robustness of samples in a validation set is a statistically weak measure. We propose instead to weight the importance of samples depending on their difficulty. We motivate the proposed score by a theoretical case study using logistic regression. We also empirically demonstrate the ability of the proposed score to measure robustness of classifiers with little dependence on the choice of samples in more complex settings, including deep convolutional neural networks and real datasets.

Transfer learning using deep neural networks as feature extractors has become increasingly popular over the past few years. It allows to obtain state-of-the-art accuracy on datasets too small to train a deep neural network on its own, and it provides cutting edge descriptors that, combined with nonparametric learning methods, allow rapid and flexible deployment of performing solutions in computationally restricted settings. In this paper, we are interested in showing that the features extracted using deep neural networks have specific properties which can be used to improve accuracy of downstream nonparametric learning methods. Namely, we demonstrate that for some distributions where information is embedded in a few coordinates, segmenting feature vectors can lead to better accuracy. We show how this model can be applied to real datasets by performing experiments using three mainstream deep neural network feature extractors and four databases, in vision and audio.

In this paper, we explore body language in mobile phones as a means of relational interaction. We describe a prototype that allows the simulation of proxemic reactions to the nearing hand of a user, ranging from affection to aversion, based on nearness-based input and shape change-based output. A user study is reported, which indicates that users were able to interpret the prototype's behavior drawing on animal parallels. It is concluded that proxemically reactive actuation may be a viable means of actively integrating the relationship between the user and the device into the interaction.

In this paper, we explore how direct physical cues of interpersonal nearness can be achieved in mobile phones. Exemplarily, we present three novel means of communication for mobile phones: grasping, kissing and whispering. Reviewing the related work, we point to a research gap in direct physical near-body actuation in mobile telecommunication. To assess this gap, we present three prototypes that implement the proposed novel means of communication. We present initial user comments on the prototypes, which point to acceptance issues. We conclude in a set of research questions for future explorations in this field.

In this paper, we present a qualitative comparison of different sketching techniques, assessing their suitability for co-designing interaction design with children. It presents a study conducted in an experimental field research, in which children aged 6-12 were engaged in a co-design process, aimed to the creation of novel communication devices or services that fit their particular needs. The study compared embodied, physical sketching (body storming that was documented as photo stories) with disembodied, drawn sketching (comics), as for their creative results, and how the children, reportedly, felt during the creation process. The results indicate that embodied sketching techniques were more suitable for the children, both as for the quality of the results, and for the subjective experience of the children while designing.

In this paper, we present a new shape-based display technique for mobile phones: A rotatory deformation of the phone's chassis, resulting in a tapering between the phone's front and back panel. It draws on proprioceptive skills of the human hand, which we hypothesize to be sensitive to paralellism and tapering of two opposing panels.

In current healthcare research, pain logs are an important means to measure the impact of medication and to detect pain patterns. However, the entry of textual pain data may be negatively reinforcing, due to its character of direct confrontation with one's pain, potentially leading to a higher pain awareness. This paper introduces valeo, a new form of a tactile pain log, which is based on a vibrotactile, pressure- and tilt-sensitive device. Future iterations will also include body part recognition. By squeezing it close to one's own body, the subjective intensity of a local pain is manifested in the device. Via a shove gesture, it is then transmitted to a docking station. Advantageously, the pain data is covert and conceptually alienated from the patient's body, which may support a possible relief. A proof-of-concept prototype and a user study indicate that our concept is perceived as convenient and reduces negative conditioning.

In this correspondence, we study a lightly loaded code-division multiple-access (CDMA) system with and without multistage hard- and soft-decision parallel interference cancelation (HD-PIC and SD-PIC). Throughout this paper we will only consider the situation of a noiseless channel, equal powers and random spreading codes. For the system with no or a fixed number of steps of interference cancelation, we give a lower bound on the maximum number of users such that the probability for the system to have no bit-errors converges to one. Moreover, we investigate when the matched filter system, where parallel interference cancelation is absent, has bit errors with probability converging to one. This implies that the use of HD-PIC and SD-PIC significantly enhances the number of users the system can serve

We introduce a form of the Hopfield model that is able to store an increasing number of biased i.i.d. patterns (it is well known that the standard Hopfield model fails to work properly in this context). We show that this new form of the Hopfield model with N neurons can store (N)/(/spl gamma/ log N) or /spl alpha/N biased patterns (depending on which notion of storage is used). The quantity /spl gamma/ increases with an increasing bias of the patterns, while /spl alpha/ decreases when the bias gets large.