Subtle adversarial image manipulations influence both human and machine perception.

Although artificial neural networks (ANNs) were inspired by the brain, ANNs exhibit a brittleness not generally observed in human perception. One shortcoming of ANNs is their susceptibility to adversarial perturbations-subtle modulations of natural images that result in changes to classification decisions, such as confidently mislabelling an image of an elephant, initially classified correctly, as a clock. In contrast, a human observer might well dismiss the perturbations as an innocuous imaging artifact. This phenomenon may point to a fundamental difference between human and machine perception, but it drives one to ask whether human sensitivity to adversarial perturbations might be revealed with appropriate behavioral measures. Here, we find that adversarial perturbations that fool ANNs similarly bias human choice. We further show that the effect is more likely driven by higher-order statistics of natural images to which both humans and ANNs are sensitive, rather than by the detailed architecture of the ANN.

Rethink reporting of evaluation results in AI.

Aggregate metrics and lack of access to results limit understanding.

Minimum and Maximum Entropy Distributions for Binary Systems with Known Means and Pairwise Correlations.

Maximum entropy models are increasingly being used to describe the collective activity of neural populations with measured mean neural activities and pairwise correlations, but the full space of probability distributions consistent with these constraints has not been explored. We provide upper and lower bounds on the entropy for the minimum entropy distribution over arbitrarily large collections of binary units with any fixed set of mean values and pairwise correlations. We also construct specific low-entropy distributions for several relevant cases. Surprisingly, the minimum entropy solution has entropy scaling logarithmically with system size for any set of first- and second-order statistics consistent with arbitrarily large systems. We further demonstrate that some sets of these low-order statistics can only be realized by small systems. Our results show how only small amounts of randomness are needed to mimic low-order statistical properties of highly entropic distributions, and we discuss some applications for engineered and biological information transmission systems.

A Device for Human Ultrasonic Echolocation.

OBJECTIVE: We present a device that combines principles of ultrasonic echolocation and spatial hearing to provide human users with environmental cues that are 1) not otherwise available to the human auditory system, and 2) richer in object and spatial information than the more heavily processed sonar cues of other assistive devices. The device consists of a wearable headset with an ultrasonic emitter and stereo microphones with affixed artificial pinnae. The goal of this study is to describe the device and evaluate the utility of the echoic information it provides. METHODS: The echoes of ultrasonic pulses were recorded and time stretched to lower their frequencies into the human auditory range, then played back to the user. We tested performance among naive and experienced sighted volunteers using a set of localization experiments, in which the locations of echo-reflective surfaces were judged using these time-stretched echoes. RESULTS: Naive subjects were able to make laterality and distance judgments, suggesting that the echoes provide innately useful information without prior training. Naive subjects were generally unable to make elevation judgments from recorded echoes. However, trained subjects demonstrated an ability to judge elevation as well. CONCLUSION: This suggests that the device can be used effectively to examine the environment and that the human auditory system can rapidly adapt to these artificial echolocation cues. SIGNIFICANCE: Interpreting and interacting with the external world constitutes a major challenge for persons who are blind or visually impaired. This device has the potential to aid blind people in interacting with their environment.

Dust deposition on the decks of the Mars Exploration Rovers: 10 years of dust dynamics on the Panoramic Camera calibration targets.

The Panoramic Cameras on NASA's Mars Exploration Rovers have each returned more than 17,000 images of their calibration targets. In order to make optimal use of this data set for reflectance calibration, a correction must be made for the presence of air fall dust. Here we present an improved dust correction procedure based on a two-layer scattering model, and we present a dust reflectance spectrum derived from long-term trends in the data set. The dust on the calibration targets appears brighter than dusty areas of the Martian surface. We derive detailed histories of dust deposition and removal revealing two distinct environments: At the Spirit landing site, half the year is dominated by dust deposition, the other half by dust removal, usually in brief, sharp events. At the Opportunity landing site the Martian year has a semiannual dust cycle with dust removal happening gradually throughout two removal seasons each year. The highest observed optical depth of settled dust on the calibration target is 1.5 on Spirit and 1.1 on Opportunity (at 601 nm). We derive a general prediction for dust deposition rates of 0.004 ± 0.001 in units of surface optical depth deposited per sol (Martian solar day) per unit atmospheric optical depth. We expect this procedure to lead to improved reflectance-calibration of the Panoramic Camera data set. In addition, it is easily adapted to similar data sets from other missions in order to deliver improved reflectance calibration as well as data on dust reflectance properties and deposition and removal history.

Modeling higher-order correlations within cortical microcolumns.

We statistically characterize the population spiking activity obtained from simultaneous recordings of neurons across all layers of a cortical microcolumn. Three types of models are compared: an Ising model which captures pairwise correlations between units, a Restricted Boltzmann Machine (RBM) which allows for modeling of higher-order correlations, and a semi-Restricted Boltzmann Machine which is a combination of Ising and RBM models. Model parameters were estimated in a fast and efficient manner using minimum probability flow, and log likelihoods were compared using annealed importance sampling. The higher-order models reveal localized activity patterns which reflect the laminar organization of neurons within a cortical column. The higher-order models also outperformed the Ising model in log-likelihood: On populations of 20 cells, the RBM had 10% higher log-likelihood (relative to an independent model) than a pairwise model, increasing to 45% gain in a larger network with 100 spatiotemporal elements, consisting of 10 neurons over 10 time steps. We further removed the need to model stimulus-induced correlations by incorporating a peri-stimulus time histogram term, in which case the higher order models continued to perform best. These results demonstrate the importance of higher-order interactions to describe the structure of correlated activity in cortical networks. Boltzmann Machines with hidden units provide a succinct and effective way to capture these dependencies without increasing the difficulty of model estimation and evaluation.

Optogenetic activation of an inhibitory network enhances feedforward functional connectivity in auditory cortex.

The mammalian neocortex is a highly interconnected network of different types of neurons organized into both layers and columns. Overlaid on this structural organization is a pattern of functional connectivity that can be rapidly and flexibly altered during behavior. Parvalbumin-positive (PV+) inhibitory neurons, which are implicated in cortical oscillations and can change neuronal selectivity, may play a pivotal role in these dynamic changes. We found that optogenetic activation of PV+ neurons in the auditory cortex enhanced feedforward functional connectivity in the putative thalamorecipient circuit and in cortical columnar circuits. In contrast, stimulation of PV+ neurons induced no change in connectivity between sites in the same layers. The activity of PV+ neurons may thus serve as a gating mechanism to enhance feedforward, but not lateral or feedback, information flow in cortical circuits. Functionally, it may preferentially enhance the contribution of bottom-up sensory inputs to perception.

Non-Gaussian statistical properties of breast images.

PURPOSE: Several studies have shown that the power spectrum of x-ray breast images is well described by a power-law at lower frequencies where anatomical variability dominates. However, an image generated from a Gaussian process with this spectrum is easily distinguished from an image of actual breast tissue by eye. This demonstrates that higher order non-Gaussian statistical properties of mammograms are readily accessible to the visual system. The authors' purpose is to quantify and characterize non-Gaussian statistical properties of breast images as influenced by processing of a digital mammogram, different imaging modalities, and breast density. METHODS: To quantify non-Gaussian statistical properties, the authors consider histograms of filter responses from the interior of a breast image that have similar properties to receptive fields in the early visual system. They quantify departure from a Gaussian distribution by the relative entropy of the histogram compared to a best-fit Gaussian distribution. This entropy is normalized by the relative entropy of a best-fit Laplacian distribution into a measure they refer to as Laplacian fractional entropy (LFE). They test the LFE on a set of 26 patients recalled at screening for which they have available full-field digital mammography (FFDM), digital breast tomosynthesis (DBT), and dedicated breast CT (bCT) images as well as breast density scores and biopsy results. RESULTS: A study of LFE in FFDM comparing the raw "for-processing" transmission data from the device to log-converted density estimates and the processed "for-display" data shows that processing mammographic image data enhances the non-Gaussian content of the image. A check of the methodology using a Gaussian process with a power-law power spectrum shows relatively little bias from the finite extent of the region of interests used. A second study comparing LFE across FFDM, DBT, and bCT modalities shows that each maximized the non-Gaussian content of the image for different ranges of spatial frequency. FFDM is optimal at high spatial frequencies (>0.7 mm(-1)), DBT is optimal at mid-range frequencies (0.3-0.7 mm(-1)), and bCT is optimal at low spatial frequency (<0.3 mm(-1)). A third study of breast density in FFDM and bCT shows that LFE generally rises slightly going from the low-to moderate density, and then falls considerably at higher densities. CONCLUSIONS: Non-Gaussian statistical structure in breast images that is manifest in the responses of Gabor filters similar to receptive fields of the early visual system is dependent on how the image data are processed, the modality used to acquire the image, and the density of the breast tissue being imaged. Higher LFE corresponds with expected improvements from image processing and 3D imaging.

New method for parameter estimation in probabilistic models: minimum probability flow.

Fitting probabilistic models to data is often difficult, due to the general intractability of the partition function. We propose a new parameter fitting method, minimum probability flow (MPF), which is applicable to any parametric model. We demonstrate parameter estimation using MPF in two cases: a continuous state space model, and an Ising spin glass. In the latter case, MPF outperforms current techniques by at least an order of magnitude in convergence time with lower error in the recovered coupling parameters.