Gal Chechik

Auditory abstraction from spectro-temporal features to coding auditory entities

The auditory system extracts behaviorally relevant information from acoustic stimuli. The average activity in auditory cortex is known to be sensitive to spectro-temporal patterns in sounds. However, it is not known whether the auditory cortex also processes more abstract features of sounds, which may be more behaviorally relevant than spectro-temporal patterns. Using recordings from three stations of the auditory pathway, the inferior colliculus (IC), the ventral division of the medial geniculate body (MGB) of the thalamus, and the primary auditory cortex (A1) of the cat in response to natural sounds, we compared the amount of information that spikes contained about two aspects of the stimuli: spectro-temporal patterns, and abstract entities present in the same stimuli such as a bird chirp, its echoes, and the ambient noise. IC spikes conveyed on average approximately the same amount of information about spectro-temporal patterns as they conveyed about abstract auditory entities, but A1 and the MGB neurons conveyed on average three times more information about abstract auditory entities than about spectro-temporal patterns. Thus, the majority of neurons in auditory thalamus and cortex coded well the presence of abstract entities in the sounds without containing much information about their spectro-temporal structure, suggesting that they are sensitive to abstract features in these sounds.

Effective Neuronal Learning with Ineffective Hebbian Learning Rules

In this article we revisit the classical neuroscience paradigm of Hebbian learning. We find that it is difficult to achieve effective associative memory storage by Hebbian synaptic learning, since it requires network-level information at the synaptic level or sparse coding level. Effective learning can yet be achieved even with nonsparse patterns by a neuronal process that maintains a zero sum of the incoming synaptic efficacies. This weight correction improves the memory capacity of associative networks from an essentially bounded one to a memory capacity that scales linearly with network size. It also enables the effective storage of patterns with multiple levels of activity within a single network. Such neuronal weight correction can be successfully carried out by activity-dependent homeostasis of the neurons synaptic efficacies, which was recently observed in cortical tissue. Thus, our findings suggest that associative learning by Hebbian synaptic learning should be accompanied by continuous remodeling of neuronally driven regulatory processes in the brain.

Learning from Noisy Large-Scale Datasets with Minimal Supervision

We present an approach to effectively use millions of images with noisy annotations in conjunction with a small subset of cleanly-annotated images to learn powerful image representations. One common approach to combine clean and noisy data is to first pre-train a network using the large noisy dataset and then fine-tune with the clean dataset. We show this approach does not fully leverage the information contained in the clean set. Thus, we demonstrate how to use the clean annotations to reduce the noise in the large dataset before fine-tuning the network using both the clean set and the full set with reduced noise. The approach comprises a multi-task network that jointly learns to clean noisy annotations and to accurately classify images. We evaluate our approach on the recently released Open Images dataset, containing ~9 million images, multiple annotations per image and over 6000 unique classes. For the small clean set of annotations we use a quarter of the validation set with ~40k images. Our results demonstrate that the proposed approach clearly outperforms direct fine-tuning across all major categories of classes in the Open Image dataset. Further, our approach is particularly effective for a large number of classes with wide range of noise in annotations (20-80% false positive annotations).

A unifying principle underlying the extracellular field potential spectral responses in the human cortex

Electrophysiological mass potentials show complex spectral changes upon neuronal activation. However, it is unknown to what extent these complex band-limited changes are interrelated or, alternatively, reflect separate neuronal processes. To address this question, intracranial electrocorticograms (ECoG) responses were recorded in patients engaged in visuomotor tasks. We found that in the 10- to 100-Hz frequency range there was a significant reduction in the exponent χ of the 1/f(χ) component of the spectrum associated with neuronal activation. In a minority of electrodes showing particularly high activations the exponent reduction was associated with specific band-limited power modulations: emergence of a high gamma (80-100 Hz) and a decrease in the alpha (9-12 Hz) peaks. Importantly, the peaks height was correlated with the 1/f(χ) exponent on activation. Control simulation ruled out the possibility that the change in 1/f(χ) exponent was a consequence of the analysis procedure. These results reveal a new global, cross-frequency (10-100 Hz) neuronal process reflected in a significant reduction of the power spectrum slope of the ECoG signal.

Specialization of Gene Expression during Mouse Brain Development

The transcriptome of the brain changes during development, reflecting processes that determine functional specialization of brain regions. We analyzed gene expression, measured using in situ hybridization across the full developing mouse brain, to quantify functional specialization of brain regions. Surprisingly, we found that during the time that the brain becomes anatomically regionalized in early development, transcription specialization actually decreases reaching a low, “neurotypic”, point around birth. This decrease of specialization is brain-wide, and mainly due to biological processes involved in constructing brain circuitry. Regional specialization rises again during post-natal development. This effect is largely due to specialization of plasticity and neural activity processes. Post-natal specialization is particularly significant in the cerebellum, whose expression signature becomes increasingly different from other brain regions. When comparing mouse and human expression patterns, the cerebellar post-natal specialization is also observed in human, but the regionalization of expression in the human Thalamus and Cortex follows a strikingly different profile than in mouse.

Localizing genes to cerebellar layers by classifying ISH images.

Gene expression controls how the brain develops and functions. Understanding control processes in the brain is particularly hard since they involve numerous types of neurons and glia, and very little is known about which genes are expressed in which cells and brain layers. Here we describe an approach to detect genes whose expression is primarily localized to a specific brain layer and apply it to the mouse cerebellum. We learn typical spatial patterns of expression from a few markers that are known to be localized to specific layers, and use these patterns to predict localization for new genes. We analyze images of in-situ hybridization (ISH) experiments, which we represent using histograms of local binary patterns (LBP) and train image classifiers and gene classifiers for four layers of the cerebellum: the Purkinje, granular, molecular and white matter layer. On held-out data, the layer classifiers achieve accuracy above 94% (AUC) by representing each image at multiple scales and by combining multiple image scores into a single gene-level decision. When applied to the full mouse genome, the classifiers predict specific layer localization for hundreds of new genes in the Purkinje and granular layers. Many genes localized to the Purkinje layer are likely to be expressed in astrocytes, and many others are involved in lipid metabolism, possibly due to the unusual size of Purkinje cells.

Context-Aware Captions from Context-Agnostic Supervision

We introduce an inference technique to produce discriminative context-aware image captions (captions that describe differences between images or visual concepts) using only generic context-agnostic training data (captions that describe a concept or an image in isolation). For example, given images and captions of siamese cat and tiger cat, we generate language that describes the siamese cat in a way that distinguishes it from tiger cat. Our key novelty is that we show how to do joint inference over a language model that is context-agnostic and a listener which distinguishes closely-related concepts. We first apply our technique to a justification task, namely to describe why an image contains a particular fine-grained category as opposed to another closely-related category of the CUB-200-2011 dataset. We then study discriminative image captioning to generate language that uniquely refers to one of two semantically-similar images in the COCO dataset. Evaluations with discriminative ground truth for justification and human studies for discriminative image captioning reveal that our approach outperforms baseline generative and speaker-listener approaches for discrimination.

Object separation in x-ray image sets

In the segmentation of natural images, most algorithms rely on the concept of occlusion. In x-ray images, however, this assumption is violated, since x-ray photons penetrate most materials. In this paper, we introduce SATISφ, a method for separating objects in a set of x-ray images using the property of additivity in log space, where the log-attenuation at a pixel is the sum of the log-attenuations of all objects that the corresponding x-ray passes through. Our method leverages multiple projection views of the same scene from slightly different angles to produce an accurate estimate of attenuation properties of objects in the scene. These properties can be used to identify the material composition of these objects, and are therefore crucial for applications like automatic threat detection. We evaluate SATISφ on a set of collected x-ray scans, showing that it outperforms a standard image segmentation approach and reduces the error of material estimation.

Gene Expression Switching of Receptor Subunits in Human Brain Development.

Synaptic receptors in the human brain consist of multiple protein subunits, many of which have multiple variants, coded by different genes, and are differentially expressed across brain regions and developmental stages. The brain can tune the electrophysiological properties of synapses to regulate plasticity and information processing by switching from one protein variant to another. Such condition-dependent variant switch during development has been demonstrated in several neurotransmitter systems including NMDA and GABA. Here we systematically detect pairs of receptor-subunit variants that switch during the lifetime of the human brain by analyzing postmortem expression data collected in a population of donors at various ages and brain regions measured using microarray and RNA-seq. To further detect variant pairs that co-vary across subjects, we present a method to quantify age-corrected expression correlation in face of strong temporal trends. This is achieved by computing the correlations in the residual expression beyond a cubic-spline model of the population temporal trend, and can be seen as a nonlinear version of partial correlations. Using these methods, we detect multiple new pairs of context dependent variants. For instance, we find a switch from GLRA2 to GLRA3 that differs from the known switch in the rat. We also detect an early switch from HTR1A to HTR5A whose trends are negatively correlated and find that their age-corrected expression is strongly positively correlated. Finally, we observe that GRIN2B switch to GRIN2A occurs mostly during embryonic development, presumably earlier than observed in rodents. These results provide a systematic map of developmental switching in the neurotransmitter systems of the human brain.

Positive correlation between ADAR expression and its targets suggests a complex regulation mediated by RNA editing in the human brain.

A-to-I RNA editing by adenosine deaminases acting on RNA is a post-transcriptional modification that is crucial for normal life and development in vertebrates. RNA editing has been shown to be very abundant in the human transcriptome, specifically at the primate-specific Alu elements. The functional role of this wide-spread effect is still not clear; it is believed that editing of transcripts is a mechanism for their down-regulation via processes such as nuclear retention or RNA degradation. Here we combine 2 neural gene expression datasets with genome-level editing information to examine the relation between the expression of ADAR genes with the expression of their target genes. Specifically, we computed the spatial correlation across structures of post-mortem human brains between ADAR and a large set of targets that were found to be edited in their Alu repeats. Surprisingly, we found that a large fraction of the edited genes are positively correlated with ADAR, opposing the assumption that editing would reduce expression. When considering the correlations between ADAR and its targets over development, 2 gene subsets emerge, positively correlated and negatively correlated with ADAR expression. Specifically, in embryonic time points, ADAR is positively correlated with many genes related to RNA processing and regulation of gene expression. These findings imply that the suggested mechanism of regulation of expression by editing is probably not a global one; ADAR expression does not have a genome wide effect reducing the expression of editing targets. It is possible, however, that RNA editing by ADAR in non-coding regions of the gene might be a part of a more complex expression regulation mechanism.