To adapt behavior to a changing environment, one must monitor outcomes of executed actions and adjust subsequent actions accordingly. Involvement of the medial frontal cortex in performance monitoring has been suggested, but little is known about neural processes that link performance monitoring to performance adjustment. Here, we recorded from neurons in the medial prefrontal cortex of monkeys learning arbitrary action-outcome contingencies. Some cells preferentially responded to positive visual feedback stimuli and others to negative feedback stimuli. The magnitude of responses to positive feedback stimuli decreased over the course of behavioral adaptation, in correlation with decreases in the amount of prediction error of action values. Therefore, these responses in medial prefrontal cells may signal the direction and amount of error in prediction of values of executed actions to specify the adjustment in subsequent action selections.

SUMMARY Knowledge about hypothetical outcomes from unchosen actions is beneficial only when such outcomes can be correctly attributed to specific actions. Here, we show that during a simulated rock-paper-scissors game, rhesus monkeys can adjust their choice behaviors according to both actual and hypothetical outcomes from their chosen and unchosen actions, respectively. In addition, neurons in both dorsolateral prefrontal cortex and orbitofrontal cortex encoded the signals related to actual and hypothetical outcomes immediately after they were revealed to the animal. Moreover, compared to the neurons in the orbitofrontal cortex, those in the dorsolateral prefrontal cortex were more likely to change their activity according to the hypothetical outcomes from specific actions. Conjunctive and parallel coding of multiple actions and their outcomes in the prefrontal cortex might enhance the efficiency of reinforcement learning and also contribute to their context-dependent memory.

Humans demonstrate behavioural advantages (biases) towards particular dimensions (colour or shape of visual objects), but such biases are significantly altered in neuropsychological disorders. Recent studies have shown that lesions in the prefrontal cortex do not abolish dimensional biases, and therefore suggest that such biases might not depend on top-down prefrontal-mediated attention and instead emerge as bottom-up processing advantages. We hypothesised that if dimensional biases merely emerge from an enhancement of object features, the presence of visual objects would be necessary for the manifestation of dimensional biases. In a specifically-designed working memory task, in which macaque monkeys and humans performed matching based on the object memory rather than the actual object, we found significant dimensional biases in both species, which appeared as a shorter response time and higher accuracy in the preferred dimension (colour and shape dimension in humans and monkeys, respectively). Moreover, the mnemonic demands of the task influenced the magnitude of dimensional bias. Our findings in two primate species indicate that the dichotomy of top-down and bottom-up processing does not fully explain the emergence of dimensional biases. Instead, dimensional biases may emerge when processed information regarding visual object features interact with mnemonic and executive functions to guide goal-directed behaviour.

Compression techniques for deep neural network models are becoming very important for the efficient execution of high-performance deep learning systems on edge-computing devices. The concept of model compression is also important for analyzing the generalization error of deep learning, known as the compression-based error bound. However, there is still huge gap between a practically effective compression method and its rigorous background of statistical learning theory. To resolve this issue, we develop a new theoretical framework for model compression and propose a new pruning method called {\it spectral pruning} based on this framework. We define the ``degrees of freedom'' to quantify the intrinsic dimensionality of a model by using the eigenvalue distribution of the covariance matrix across the internal nodes and show that the compression ability is essentially controlled by this quantity. Moreover, we present a sharp generalization error bound of the compressed model and characterize the bias--variance tradeoff induced by the compression procedure. We apply our method to several datasets to justify our theoretical analyses and show the superiority of the the proposed method.