Attention fine-tunes auditory-motor processing of speech sounds.

The earliest stages of cortical processing of speech sounds take place in the auditory cortex. Transcranial magnetic stimulation (TMS) studies have provided evidence that the human articulatory motor cortex contributes also to speech processing. For example, stimulation of the motor lip representation influences specifically discrimination of lip-articulated speech sounds. However, the timing of the neural mechanisms underlying these articulator-specific motor contributions to speech processing is unknown. Furthermore, it is unclear whether they depend on attention. Here, we used magnetoencephalography and TMS to investigate the effect of attention on specificity and timing of interactions between the auditory and motor cortex during processing of speech sounds. We found that TMS-induced disruption of the motor lip representation modulated specifically the early auditory-cortex responses to lip-articulated speech sounds when they were attended. These articulator-specific modulations were left-lateralized and remarkably early, occurring 60-100 ms after sound onset. When speech sounds were ignored, the effect of this motor disruption on auditory-cortex responses was nonspecific and bilateral, and it started later, 170 ms after sound onset. The findings indicate that articulatory motor cortex can contribute to auditory processing of speech sounds even in the absence of behavioral tasks and when the sounds are not in the focus of attention. Importantly, the findings also show that attention can selectively facilitate the interaction of the auditory cortex with specific articulator representations during speech processing.

Three types of incremental learning.

Incrementally learning new information from a non-stationary stream of data, referred to as 'continual learning', is a key feature of natural intelligence, but a challenging problem for deep neural networks. In recent years, numerous deep learning methods for continual learning have been proposed, but comparing their performances is difficult due to the lack of a common framework. To help address this, we describe three fundamental types, or 'scenarios', of continual learning: task-incremental, domain-incremental and class-incremental learning. Each of these scenarios has its own set of challenges. To illustrate this, we provide a comprehensive empirical comparison of currently used continual learning strategies, by performing the Split MNIST and Split CIFAR-100 protocols according to each scenario. We demonstrate substantial differences between the three scenarios in terms of difficulty and in terms of the effectiveness of different strategies. The proposed categorization aims to structure the continual learning field, by forming a key foundation for clearly defining benchmark problems.

Brain-inspired replay for continual learning with artificial neural networks.

Artificial neural networks suffer from catastrophic forgetting. Unlike humans, when these networks are trained on something new, they rapidly forget what was learned before. In the brain, a mechanism thought to be important for protecting memories is the reactivation of neuronal activity patterns representing those memories. In artificial neural networks, such memory replay can be implemented as 'generative replay', which can successfully - and surprisingly efficiently - prevent catastrophic forgetting on toy examples even in a class-incremental learning scenario. However, scaling up generative replay to complicated problems with many tasks or complex inputs is challenging. We propose a new, brain-inspired variant of replay in which internal or hidden representations are replayed that are generated by the network's own, context-modulated feedback connections. Our method achieves state-of-the-art performance on challenging continual learning benchmarks (e.g., class-incremental learning on CIFAR-100) without storing data, and it provides a novel model for replay in the brain.

Parsing Hippocampal Theta Oscillations by Nested Spectral Components during Spatial Exploration and Memory-Guided Behavior.

Theta oscillations reflect rhythmic inputs that continuously converge to the hippocampus during exploratory and memory-guided behavior. The theta-nested operations that organize hippocampal spiking could either occur regularly from one cycle to the next or be tuned on a cycle-by-cycle basis. To resolve this, we identified spectral components nested in individual theta cycles recorded from the mouse CA1 hippocampus. Our single-cycle profiling revealed theta spectral components associated with different firing modulations and distinguishable ensembles of principal cells. Moreover, novel co-firing patterns of principal cells in theta cycles nesting mid-gamma oscillations were the most strongly reactivated in subsequent offline sharp-wave/ripple events. Finally, theta-nested spectral components were differentially altered by behavioral stages of a memory task; the 80-Hz mid-gamma component was strengthened during learning, whereas the 22-Hz beta, 35-Hz slow gamma, and 54-Hz mid-gamma components increased during retrieval. We conclude that cycle-to-cycle variability of theta-nested spectral components allows parsing of theta oscillations into transient operating modes with complementary mnemonic roles.

Hippocampal Offline Reactivation Consolidates Recently Formed Cell Assembly Patterns during Sharp Wave-Ripples.

The ability to reinstate neuronal assemblies representing mnemonic information is thought to require their consolidation through offline reactivation during sleep/rest. To test this, we detected cell assembly patterns formed by repeated neuronal co-activations in the mouse hippocampus during exploration of spatial environments. We found that the reinstatement of assembly patterns representing a novel, but not a familiar, environment correlated with their offline reactivation and was impaired by closed-loop optogenetic disruption of sharp wave-ripple oscillations. Moreover, we discovered that reactivation was only required for the reinstatement of assembly patterns whose expression was gradually strengthened during encoding of a novel place. The context-dependent reinstatement of assembly patterns whose expression did not gain in strength beyond the first few minutes of spatial encoding was not dependent on reactivation. This demonstrates that the hippocampus can hold concurrent representations of space that markedly differ in their encoding dynamics and their dependence on offline reactivation for consolidation. VIDEO ABSTRACT.

Recoding a cocaine-place memory engram to a neutral engram in the hippocampus.

The hippocampus provides the brain's memory system with a subset of neurons holding a map-like representation of each environment experienced. We found in mice that optogenetic silencing those neurons active in an environment unmasked a subset of quiet neurons, enabling the emergence of an alternative map. When applied in a cocaine-paired environment, this intervention neutralized an otherwise long-lasting drug-place preference, showing that recoding a spatial memory engram can alleviate associated maladaptive behavior.