Cognition of Time and Thinking Beyond.

A common research protocol in cognitive neuroscience is to train subjects to perform deliberately designed experiments while recording brain activity, with the aim of understanding the brain mechanisms underlying cognition. However, how the results of this protocol of research can be applied in technology is seldom discussed. Here, I review the studies on time processing of the brain as examples of this research protocol, as well as two main application areas of neuroscience (neuroengineering and brain-inspired artificial intelligence). Time processing is a fundamental dimension of cognition, and time is also an indispensable dimension of any real-world signal to be processed in technology. Therefore, one may expect that the studies of time processing in cognition profoundly influence brain-related technology. Surprisingly, I found that the results from cognitive studies on timing processing are hardly helpful in solving practical problems. This awkward situation may be due to the lack of generalizability of the results of cognitive studies, which are under well-controlled laboratory conditions, to real-life situations. This lack of generalizability may be rooted in the fundamental unknowability of the world (including cognition). Overall, this paper questions and criticizes the usefulness and prospect of the abovementioned research protocol of cognitive neuroscience. I then give three suggestions for future research. First, to improve the generalizability of research, it is better to study brain activity under real-life conditions instead of in well-controlled laboratory experiments. Second, to overcome the unknowability of the world, we can engineer an easily accessible surrogate of the object under investigation, so that we can predict the behavior of the object under investigation by experimenting on the surrogate. Third, the paper calls for technology-oriented research, with the aim of technology creation instead of knowledge discovery.

Top-down generation of low-resolution representations improves visual perception and imagination.

Perception or imagination requires top-down signals from high-level cortex to primary visual cortex (V1) to reconstruct or simulate the representations bottom-up stimulated by the seen images. Interestingly, top-down signals in V1 have lower spatial resolution than bottom-up representations. It is unclear why the brain uses low-resolution signals to reconstruct or simulate high-resolution representations. By modeling the top-down pathway of the visual system using the decoder of a variational auto-encoder (VAE), we reveal that low-resolution top-down signals can better reconstruct or simulate the information contained in the sparse activities of V1 simple cells, which facilitates perception and imagination. This advantage of low-resolution generation is related to facilitating high-level cortex to form geometry-respecting representations observed in experiments. Furthermore, we present two findings regarding this phenomenon in the context of AI-generated sketches, a style of drawings made of lines. First, we found that the quality of the generated sketches critically depends on the thickness of the lines in the sketches: thin-line sketches are harder to generate than thick-line sketches. Second, we propose a technique to generate high-quality thin-line sketches: instead of directly using original thin-line sketches, we use blurred sketches to train VAE or GAN (generative adversarial network), and then infer the thin-line sketches from the VAE- or GAN-generated blurred sketches. Collectively, our work suggests that low-resolution top-down generation is a strategy the brain uses to improve visual perception and imagination, which inspires new sketch-generation AI techniques.

Visual cortex encodes timing information in humans and mice.

Despite the importance of timing in our daily lives, our understanding of how the human brain mediates second-scale time perception is limited. Here, we combined intracranial stereoelectroencephalography (SEEG) recordings in epileptic patients and circuit dissection in mice to show that visual cortex (VC) encodes timing information. We first asked human participants to perform an interval-timing task and found VC to be a key timing brain area. We then conducted optogenetic experiments in mice and showed that VC plays an important role in the interval-timing behavior. We further found that VC neurons fired in a time-keeping sequential manner and exhibited increased excitability in a timed manner. Finally, we used a computational model to illustrate a self-correcting learning process that generates interval-timed activities with scalar-timing property. Our work reveals how localized oscillations in VC occurring in the seconds to deca-seconds range relate timing information from the external world to guide behavior.

Understanding the computation of time using neural network models.

To maximize future rewards in this ever-changing world, animals must be able to discover the temporal structure of stimuli and then anticipate or act correctly at the right time. How do animals perceive, maintain, and use time intervals ranging from hundreds of milliseconds to multiseconds in working memory? How is temporal information processed concurrently with spatial information and decision making? Why are there strong neuronal temporal signals in tasks in which temporal information is not required? A systematic understanding of the underlying neural mechanisms is still lacking. Here, we addressed these problems using supervised training of recurrent neural network models. We revealed that neural networks perceive elapsed time through state evolution along stereotypical trajectory, maintain time intervals in working memory in the monotonic increase or decrease of the firing rates of interval-tuned neurons, and compare or produce time intervals by scaling state evolution speed. Temporal and nontemporal information is coded in subspaces orthogonal with each other, and the state trajectories with time at different nontemporal information are quasiparallel and isomorphic. Such coding geometry facilitates the decoding generalizability of temporal and nontemporal information across each other. The network structure exhibits multiple feedforward sequences that mutually excite or inhibit depending on whether their preferences of nontemporal information are similar or not. We identified four factors that facilitate strong temporal signals in nontiming tasks, including the anticipation of coming events. Our work discloses fundamental computational principles of temporal processing, and it is supported by and gives predictions to a number of experimental phenomena.

Spike Pattern Structure Influences Synaptic Efficacy Variability under STDP and Synaptic Homeostasis. II: Spike Shuffling Methods on LIF Networks.

Synapses may undergo variable changes during plasticity because of the variability of spike patterns such as temporal stochasticity and spatial randomness. Here, we call the variability of synaptic weight changes during plasticity to be efficacy variability. In this paper, we investigate how four aspects of spike pattern statistics (i.e., synchronous firing, burstiness/regularity, heterogeneity of rates and heterogeneity of cross-correlations) influence the efficacy variability under pair-wise additive spike-timing dependent plasticity (STDP) and synaptic homeostasis (the mean strength of plastic synapses into a neuron is bounded), by implementing spike shuffling methods onto spike patterns self-organized by a network of excitatory and inhibitory leaky integrate-and-fire (LIF) neurons. With the increase of the decay time scale of the inhibitory synaptic currents, the LIF network undergoes a transition from asynchronous state to weak synchronous state and then to synchronous bursting state. We first shuffle these spike patterns using a variety of methods, each designed to evidently change a specific pattern statistics; and then investigate the change of efficacy variability of the synapses under STDP and synaptic homeostasis, when the neurons in the network fire according to the spike patterns before and after being treated by a shuffling method. In this way, we can understand how the change of pattern statistics may cause the change of efficacy variability. Our results are consistent with those of our previous study which implements spike-generating models on converging motifs. We also find that burstiness/regularity is important to determine the efficacy variability under asynchronous states, while heterogeneity of cross-correlations is the main factor to cause efficacy variability when the network moves into synchronous bursting states (the states observed in epilepsy).

Spike Pattern Structure Influences Synaptic Efficacy Variability under STDP and Synaptic Homeostasis. I: Spike Generating Models on Converging Motifs.

In neural systems, synaptic plasticity is usually driven by spike trains. Due to the inherent noises of neurons and synapses as well as the randomness of connection details, spike trains typically exhibit variability such as spatial randomness and temporal stochasticity, resulting in variability of synaptic changes under plasticity, which we call efficacy variability. How the variability of spike trains influences the efficacy variability of synapses remains unclear. In this paper, we try to understand this influence under pair-wise additive spike-timing dependent plasticity (STDP) when the mean strength of plastic synapses into a neuron is bounded (synaptic homeostasis). Specifically, we systematically study, analytically and numerically, how four aspects of statistical features, i.e., synchronous firing, burstiness/regularity, heterogeneity of rates and heterogeneity of cross-correlations, as well as their interactions influence the efficacy variability in converging motifs (simple networks in which one neuron receives from many other neurons). Neurons (including the post-synaptic neuron) in a converging motif generate spikes according to statistical models with tunable parameters. In this way, we can explicitly control the statistics of the spike patterns, and investigate their influence onto the efficacy variability, without worrying about the feedback from synaptic changes onto the dynamics of the post-synaptic neuron. We separate efficacy variability into two parts: the drift part (DriftV) induced by the heterogeneity of change rates of different synapses, and the diffusion part (DiffV) induced by weight diffusion caused by stochasticity of spike trains. Our main findings are: (1) synchronous firing and burstiness tend to increase DiffV, (2) heterogeneity of rates induces DriftV when potentiation and depression in STDP are not balanced, and (3) heterogeneity of cross-correlations induces DriftV together with heterogeneity of rates. We anticipate our work important for understanding functional processes of neuronal networks (such as memory) and neural development.

Optimal cooperation-trap strategies for the iterated rock-paper-scissors game.

In an iterated non-cooperative game, if all the players act to maximize their individual accumulated payoff, the system as a whole usually converges to a Nash equilibrium that poorly benefits any player. Here we show that such an undesirable destiny is avoidable in an iterated Rock-Paper-Scissors (RPS) game involving two rational players, X and Y. Player X has the option of proactively adopting a cooperation-trap strategy, which enforces complete cooperation from the rational player Y and leads to a highly beneficial and maximally fair situation to both players. That maximal degree of cooperation is achievable in such a competitive system with cyclic dominance of actions may stimulate further theoretical and empirical studies on how to resolve conflicts and enhance cooperation in human societies.