Deep neural networks (DNNs) have had extraordinary successes in classifying photographic images of objects and are often described as the best models of biological vision. This conclusion is largely based on three sets of findings: (1) DNNs are more accurate than any other model in classifying images taken from various datasets, (2) DNNs do the best job in predicting the pattern of human errors in classifying objects taken from various behavioral datasets, and (3) DNNs do the best job in predicting brain signals in response to images taken from various brain datasets (e.g., single cell responses or fMRI data). However, these behavioral and brain datasets do not test hypotheses regarding what features are contributing to good predictions and we show that the predictions may be mediated by DNNs that share little overlap with biological vision. More problematically, we show that DNNs account for almost no results from psychological research. This contradicts the common claim that DNNs are good, let alone the best, models of human object recognition. We argue that theorists interested in developing biologically plausible models of human vision need to direct their attention to explaining psychological findings. More generally, theorists need to build models that explain the results of experiments that manipulate independent variables designed to test hypotheses rather than compete on making the best predictions. We conclude by briefly summarizing various promising modelling approaches that focus on psychological data.

As modern mobile devices increase in their capability and accessibility, they introduce additional demands in terms of security - particularly authentication. With the widely documented poor use of PINs, Active Authentication is designed to overcome the fundamental issue of usable and secure authentication through utilizing biometric-based techniques to continuously verify user identity. This paper proposes a novel text-based multimodal biometric approach utilizing linguistic analysis, keystroke dynamics and behavioural profiling. Experimental investigations show that users can be discriminated via their text-based entry, with an average Equal Error Rate (EER) of 3.3%. Based on these findings, a framework that is able to provide robust, continuous and transparent authentication is proposed. The framework is evaluated to examine the effectiveness of providing security and user convenience. The result showed that the framework is able to provide a 91% reduction in the number of intrusive authentication requests required for high security applications

Visual translation tolerance refers to our capacity to recognize objects over a wide range of different retinal locations. Although translation is perhaps the simplest spatial transform that the visual system needs to cope with, the extent to which the human visual system can identify objects at previously unseen locations is unclear, with some studies reporting near complete invariance over 10 degrees and other reporting zero invariance at 4 degrees of visual angle. Similarly, there is confusion regarding the extent of translation tolerance in computational models of vision, as well as the degree of match between human and model performance. Here, we report a series of eye-tracking studies (total N = 70) demonstrating that novel objects trained at one retinal location can be recognized at high accuracy rates following translations up to 18 degrees. We also show that standard deep convolutional neural networks (DCNNs) support our findings when pretrained to classify another set of stimuli across a range of locations, or when a global average pooling (GAP) layer is added to produce larger receptive fields. Our findings provide a strong constraint for theories of human vision and help explain inconsistent findings previously reported with convolutional neural networks (CNNs).

No abstract

Over the last 150 years, human manual reaction times (RTs) have been recorded countless times. Yet, our understanding of them remains remarkably poor. RTs are highly variable with positively skewed frequency distributions, often modeled as an inverse Gaussian distribution reflecting a stochastic rise to threshold (diffusion process). However, latency distributions of saccades are very close to the reciprocal Normal, suggesting that “rate” (reciprocal RT) may be the more fundamental variable. We explored whether this phenomenon extends to choice manual RTs. We recorded two-alternative choice RTs from 24 subjects, each with 4 blocks of 200 trials with two task difficulties (easy vs. difficult discrimination) and two instruction sets (urgent vs. accurate). We found that rate distributions were, indeed, very close to Normal, shifting to lower rates with increasing difficulty and accuracy, and for some blocks they appeared to become left-truncated, but still close to Normal. Using autoregressive techniques, we found temporal sequential dependencies for lags of at least 3. We identified a transient and steady-state component in each block. Because rates were Normal, we were able to estimate autoregressive weights using the Box-Jenkins technique, and convert to a moving average model using z-transforms to show explicit dependence on stimulus input. We also found a spatial sequential dependence for the previous 3 lags depending on whether the laterality of previous trials was repeated or alternated. This was partially dissociated from temporal dependency as it only occurred in the easy tasks. We conclude that 2-alternative choice manual RT distributions are close to reciprocal Normal and not the inverse Gaussian. This is not consistent with stochastic rise to threshold models, and we propose a simple optimality model in which reward is maximized to yield to an optimal rate, and hence an optimal time to respond. We discuss how it might be implemented.

No abstract

Deep neural networks (DNNs) have had extraordinary successes in classifying photographic images of objects and are often described as the best models of biological vision. This conclusion is largely based on three sets of findings: (1) DNNs are more accurate than any other model in classifying images taken from various datasets, (2) DNNs do the best job in predicting the pattern of human errors in classifying objects taken from various behavioral benchmark datasets, and (3) DNNs do the best job in predicting brain signals in response to images taken from various brain benchmark datasets (e.g., single cell responses or fMRI data). However, most behavioral and brain benchmarks report the outcomes of observational experiments that do not manipulate any independent variables, and we show that the good prediction on these datasets may be mediated by DNNs that share little overlap with biological vision. More problematically, we show that DNNs account for almost no results from psychological research. This contradicts the common claim that DNNs are good, let alone the best, models of human object recognition. We argue that theorists interested in developing biologically plausible models of human vision need to direct their attention to explaining psychological findings. More generally, theorists need to build models that explain the results of experiments that manipulate independent variables designed to test hypotheses rather than compete on predicting observational data. We conclude by briefly summarizing various promising modelling approaches that focus on psychological data.

Part 3: Mobile SecurityInternational audienceAs modern mobile devices are increasing in their capability and accessibility, they introduce additional demands in terms of security - particularly authentication. With the widely documented poor use of PINs, Active Authentication is designed to overcome the fundamental issue of usable and secure authentication through utilizing biometric-based techniques to continuously verify user identity. This paper proposes a novel text-based multimodal biometric approach utilizing linguistic analysis, keystroke dynamics and behavioral profiling. Experimental investigations show that users can be discriminated via their text-based entry, with an average Equal Error Rate (EER) of 3.3%. Based on these findings, a framework that is able to provide robust, continuous and transparent authentication is proposed. The framework is evaluated to examine the effectiveness of providing security and user convenience. The result showed that the framework is able to provide a 91% reduction in the number of intrusive authentication requests required for high security applications