The Kavli Foundation

Visual acuity is better for vertical and horizontal compared to other orientations. This cross-species phenomenon is often explained by “efficient coding,” whereby more neurons show sharper tuning for the orientations most common in natural vision. However, it is unclear if experience alone can account for such biases. Here, we measured orientation representations in a convolutional neural network, VGG-16, trained on modified versions of ImageNet (rotated by 0°, 22.5°, or 45° counterclockwise of upright). Discriminability for each model was highest near the orientations that were most common in the network's training set. Furthermore, there was an overrepresentation of narrowly tuned units selective for the most common orientations. These effects emerged in middle layers and increased with depth in the network, though this layer-wise pattern may depend on properties of the evaluation stimuli used. Biases emerged early in training, consistent with the possibility that nonuniform representations may play a functional role in the network's task performance. Together, our results suggest that biased orientation representations can emerge through experience with a nonuniform distribution of orientations, supporting the efficient coding hypothesis.

Working memory (WM) allows observers to hold information in mind to eventually guide behavior. In some cases, information is stored in a sensory-like format so the details of an object can later be distinguished from similar items. In other cases, it is advantageous to use a more motor-like coding scheme, for example, instead of remembering the visual details of your route to work, you remember a series of motor commands. Here, we used fMRI and an inverted encoding model to compare feature representations under conditions encouraging sensory-like or motor-like strategies. Subjects remembered the orientation of a briefly (500ms) presented grating over a variable delay (2–8s). Next, a spinning dial was shown, rotating at a fixed speed for 3s. Subjects pressed a button when the dial matched the remembered orientation; no feedback was given. We manipulated the predictability of the dial starting position and rotation direction. On trials where both were known beforehand, subjects could plan their motor response as soon as the grating appeared, while on trials where one or both were unknown, subjects could not directly divert to a motor plan. We found that during the fully predictable condition, representations in primary motor cortex were poorly modeled by a sensory training set, but were robust when trained and tested within the memory task itself. The other three conditions showed the opposite effect, suggesting the neural code was less sensory-like when responses could be planned in advance. This effect developed gradually along the posterior-anterior axis of the brain. Additionally, when rotation direction was predictable, both neural reconstructions and behavioral responses were biased away from the expected rotation direction, suggesting that subjects were attempting to compensate for their response latency. Overall, these findings demonstrate that the brain can employ multiple complementary strategies for encoding information that are not mutually exclusive.

When viewing familiar stimuli (e.g., common words), processing is highly automatized such that it can interfere with the processing of incompatible sensory information. At least two mechanisms may help mitigate this interference. Early selection accounts posit that attentional processes filter out distracting sensory information to avoid conflict. Alternatively, late selection accounts hold that all sensory inputs receive full semantic analysis and that frontal executive mechanisms are recruited to resolve conflict. To test how these mechanisms operate to overcome conflict induced by highly automatized processing, we developed a novel version of the color-word Stroop task, where targets and distractors were simultaneously flickered at different frequencies. We measured the quality of early sensory processing by assessing the amplitude of steady-state visually evoked potentials (SSVEPs) elicited by targets and distractors. We also indexed frontal executive processes by assessing changes in frontal theta oscillations induced by color-word incongruency. We found that target- and distractor-related SSVEPs were not modulated by changes in the level of conflict whereas frontal theta activity increased on high compared to low conflict trials. These results suggest that frontal executive processes play a more dominant role in mitigating cognitive interference driven by the automatic tendency to process highly familiar stimuli.

The microbiome presents great opportunities for understanding and improving the world around us and elucidating the interactions that compose it. The microbiome also poses tremendous challenges for mapping and manipulating the entangled networks of interactions among myriad diverse organisms. Here, we describe the opportunities, technical needs, and potential approaches to address these challenges, based on recent and upcoming advances in measurement and control at the nanoscale and beyond. These technical needs will provide the basis for advancing the largely descriptive studies of the microbiome to the theoretical and mechanistic understandings that will underpin the discipline of microbiome engineering. We anticipate that the new tools and methods developed will also be more broadly useful in environmental monitoring, medicine, forensics, and other areas.