The AMAAI lab is engaged in cutting-edge research at the intersection of music, audio, and artificial intelligence. Our PhD students contribute to groundbreaking projects that explore areas such as Generative Music AI, Music Information Retrieval, AI Music Production, and Affective Computing for Music. A PhD in the AMAAI lab offers the opportunity to conduct research at the forefront of a rapidly developing field, gain experience in presenting research at top academic conferences, publishing papers in prestigious journals, and potentially forge collaborations with leading figures in the music industry.

Recent advances in artificial intelligence have led to new forms of misinformation, including highly realistic “deepfake” synthetic media. We conduct three experiments to investigate how and why retail investors react to deepfake financial news. Results from the first two experiments provide evidence that investors use a “realism heuristic,” responding more intensely to audio and video deepfakes as their perceptual realism increases. In the third experiment, we introduce an intervention to prompt analytical thinking, varying whether participants make analytical judgments about credibility or intuitive investment judgments. When making intuitive investment judgments, investors are strongly influenced by both more and less realistic deepfakes. When making analytical credibility judgments, investors are able to discern the non-credibility of less realistic deepfakes but struggle with more realistic deepfakes. Thus, while analytical thinking can reduce the impact of less realistic deepfakes, highly realistic deepfakes are able to overcome this analytical scrutiny. Our results suggest that deepfake financial news poses novel threats to investors.

Subsecond timing underlies nearly all sensory and motor activities across species and is critical to survival. While subsecond temporal information has been found across cortical and subcortical regions, it is unclear if it is generated locally and intrinsically or if it is a read out of a centralized clock-like mechanism. Indeed, mechanisms of subsecond timing at the circuit level are largely obscure. Primary sensory areas are well-suited to address these question as they have early access to sensory information and provide minimal processing to it: if temporal information is found in these regions, it is likely to be generated intrinsically and locally. We test this hypothesis by training mice to perform an audio-visual temporal pattern sensory discrimination task as we use 2-photon calcium imaging, a technique capable of recording population level activity at single cell resolution, to record activity in primary visual cortex (V1). We have found significant changes in network dynamics through mice’s learning of the task from naive to middle to expert levels. Changes in network dynamics and behavioral performance are well accounted for by an intrinsic model of timing in which the trajectory of q network through high dimensional state space represents temporal sensory information. Conversely, while we found evidence of other temporal encoding models, such as oscillatory activity, we did not find that they accounted for increased performance but were in fact correlated with the intrinsic model itself. These results provide insight into how subsecond temporal information is encoded mechanistically at the circuit level.

Young children have sophisticated representations of their visual and linguistic environment. Where do these representations come from? How much knowledge arises through generic learning mechanisms applied to sensory data, and how much requires more substantive (possibly innate) inductive biases? We examine these questions by training neural networks solely on longitudinal data collected from a single child (Sullivan et al., 2020), consisting of egocentric video and audio streams. Our principal findings are as follows: 1) Based on visual only training, neural networks can acquire high-level visual features that are broadly useful across categorization and segmentation tasks. 2) Based on language only training, networks can acquire meaningful clusters of words and sentence-level syntactic sensitivity. 3) Based on paired visual and language training, networks can acquire word-referent mappings from tens of noisy examples and align their multi-modal conceptual systems. Taken together, our results show how sophisticated visual and linguistic representations can arise through data-driven learning applied to one child’s first-person experience.

Memory is a key component of biological neural systems that enables the retention of information over a huge range of temporal scales, ranging from hundreds of milliseconds up to years. While Hebbian plasticity is believed to play a pivotal role in biological memory, it has so far been analyzed mostly in the context of pattern completion and unsupervised learning. Here, we propose that Hebbian plasticity is fundamental for computations in biological neural systems. We introduce a novel spiking neural network (SNN) architecture that is enriched by Hebbian synaptic plasticity. We experimentally show that our memory-equipped SNN model outperforms state-of-the-art deep learning mechanisms in a sequential pattern-memorization task, as well as demonstrate superior out-of-distribution generalization capabilities compared to these models. We further show that our model can be successfully applied to one-shot learning and classification of handwritten characters, improving over the state-of-the-art SNN model. We also demonstrate the capability of our model to learn associations for audio to image synthesis from spoken and handwritten digits. Our SNN model further presents a novel solution to a variety of cognitive question answering tasks from a standard benchmark, achieving comparable performance to both memory-augmented ANN and SNN-based state-of-the-art solutions to this problem. Finally we demonstrate that our model is able to learn from rewards on an episodic reinforcement learning task and attain near-optimal strategy on a memory-based card game. Hence, our results show that Hebbian enrichment renders spiking neural networks surprisingly versatile in terms of their computational as well as learning capabilities. Since local Hebbian plasticity can easily be implemented in neuromorphic hardware, this also suggests that powerful cognitive neuromorphic systems can be build based on this principle.

Our ability to perceive the world around us can be affected by a number of factors including the nature of the external information, prior experience of the environment, and the integrity of the underlying perceptual system. A particular challenge for the brain is to maintain a coherent perception from information encoded by the peripheral sensory organs whose function is affected by typical, developmental changes across the lifespan. Yet, how the brain adapts to the maturation of the senses, as well as experiential changes in the multisensory environment, is poorly understood. Over the past few years, we have used a range of multisensory tasks to investigate the role of ageing on the brain’s ability to merge sensory inputs. In particular, we have embedded an audio-visual task based on the sound-induced flash illusion (SIFI) into a large-scale, longitudinal study of ageing. Our findings support the idea that the temporal binding window (TBW) is modulated by age and reveal important individual differences in this TBW that may have clinical implications. However, our investigations also suggest the TWB is experience-dependent with evidence for both long and short term behavioural plasticity. An overview of these findings, including recent evidence on how multisensory integration may be associated with higher order functions, will be discussed.

Multisensory responses are ubiquitous in so-called unisensory cortex. However, despite their prevalence, we have very little understanding of what – if anything - they contribute to perception. In this talk I will focus on audio-visual integration in auditory cortex. Anatomical tracing studies highlight visual cortex as one source of visual input to auditory cortex. Using cortical cooling we test the hypothesis that these inputs support audiovisual integration in ferret auditory cortex. Behavioural studies in humans support the idea that visual stimuli can help listeners to parse an auditory scene. This effect is paralleled in single units in auditory cortex, where responses to a sound mixture can be determined by the timing of a visual stimulus such that sounds that are temporally coherent with a visual stimulus are preferentially represented. Our recent data therefore support the idea that one role for the early integration of auditory and visual signals in auditory cortex is to support auditory scene analysis, and that visual cortex plays a key role in this process.

Social interactions such as mating and fighting are driven by internal emotional states. How can we study internal states of an animal when it cannot tell us its subjective feelings? Especially when the meaning of the animal’s behavior is not clear to us, can we understand the underlying internal states of the animal? In this talk, I will introduce our recent work in which we used male mounting behavior in mice as an example to understand the underlying internal state of the animals. In many animal species, males exhibit mounting behavior toward females as part of the mating behavior repertoire. Interestingly, males also frequently show mounting behavior toward other males of the same species. It is not clear what the underlying motivation is - whether it is reproductive in nature or something distinct. Through detailed analysis of video and audio recordings during social interactions, we found that while male-directed and female-directed mounting behaviors are motorically similar, they can be distinguished by both the presence of ultrasonic vocalization during female-directed mounting (reproductive mounting) and the display of aggression following male-directed mounting (aggressive mounting). Using optogenetics, we further identified genetically defined neural populations in the medial preoptic area (MPOA) that mediate reproductive mounting and the ventrolateral ventromedial hypothalamus (VMHvl) that mediate aggressive mounting. In vivo microendocsopic imaging in MPOA and VMHvl revealed distinct neural ensembles that mainly encode either a reproductive or an aggressive state during which male or female directed mounting occurs. Together, these findings demonstrate that internal states are represented in the hypothalamus and that motorically similar behaviors exhibited under different contexts may reflect distinct internal states.

Team work makes sensory streams work: our senses work together, learn from each other, and stand in for one another, the result of which is perception and understanding. Learned associations between stimuli in different sensory modalities can shape the way we perceive these stimuli (Mcgurk and Macdonald, 1976). During audio-visual associative learning, auditory cortex is thought to underlie multi-modal plasticity in visual cortex (McIntosh et al., 1998; Mishra et al., 2007; Zangenehpour and Zatorre, 2010). However, it is not well understood how processing in visual cortex is altered by an auditory stimulus that is predictive of a visual stimulus and what the mechanisms are that mediate such experience-dependent, audio-visual associations in sensory cortex. Here we describe a neural mechanism by which an auditory input can shape visual representations of behaviorally relevant stimuli through direct interactions between auditory and visual cortices. We show that the association of an auditory stimulus with a visual stimulus in a behaviorally relevant context leads to an experience-dependent suppression of visual responses in primary visual cortex (V1). Auditory cortex axons carry a mixture of auditory and retinotopically-matched visual input to V1, and optogenetic stimulation of these axons selectively suppresses V1 neurons responsive to the associated visual stimulus after, but not before, learning. Our results suggest that cross-modal associations can be stored in long-range cortical connections and that with learning these cross-modal connections function to suppress the responses to predictable input.

What do other people really think of you? In this talk, I highlight the unique perspective that other people have on the most consequential aspects of our personalities—how we treat others, our best and worst qualities, and our moral character. First, I compare how people thought they behaved with how they actually behaved in everyday life (based on observer ratings of unobtrusive audio recordings; 217 people, 2,519 observations). I show that when people think they are being kind (vs. rude), others do not necessarily agree. This suggests that people may have blind spots about how well they are treating others in the moment. Next, I compare what 463 people thought their own best and worst traits were with what their friends thought about them. The results reveal that friends are more likely to point out flaws in the prosocial and moral domains (e.g., “inconsiderate”, “selfish”, “manipulative”) than are people themselves. Does this imply that others might want us to be more moral? To find out, I compare what targets (N = 800) want to change about their own personalities with what their close others (N = 958) want to change about them. The results show that people don’t particularly want to be more moral, and their close others don’t want them to be more moral, either. I conclude with future directions on honest feedback as a pathway to self-insight and, ultimately, self-improvement.

Back-propagation is a powerful supervised learning algorithm in artificial neural networks, because it solves the credit assignment problem (essentially: what should the hidden layers do?). This algorithm has led to the deep learning revolution. But unfortunately, back-propagation cannot be used directly in spiking neural networks (SNN). Indeed, it requires differentiable activation functions, whereas spikes are all-or-none events which cause discontinuities. Here we present two strategies to overcome this problem. The first one is to use a so-called 'surrogate gradient', that is to approximate the derivative of the threshold function with the derivative of a sigmoid. We will present some applications of this method for time series processing (audio, internet traffic, EEG). The second one concerns a specific class of SNNs, which process static inputs using latency coding with at most one spike per neuron. Using approximations, we derived a latency-based back-propagation rule for this sort of networks, called S4NN, and applied it to image classification.