Abstract: Disclosed are systems, methods, and non-transitory media for performing radio frequency sensing detection operations. For instance, radio frequency data can be received that is associated with at least one wireless device. The radio frequency data can be based on radio frequency signals reflected from a first object and received by the at least one wireless device. Training label data can also be obtained (e.g., from a labeling device, from the at least one wireless device, etc.). The training label data can be based at least in part on the first object and input data (e.g., received by the labeling device, the at least one wireless device, etc.). A sensing model can be generated based on the radio frequency data and the training label data.

Abstract: An apparatus may utilize an air interface to transmit and/or receive a transmission during a first TTI that includes a second set of data overriding a first set of data scheduled for transmission during the first TTI. The air interface may further be utilized to transmit and/or receive a transmission after a duration of time that is less than a total time duration of the first TTI that includes an override indicator that is at least partially embedded in a data portion of a subframe. The override indicator may be configured to indicate that the first set of data scheduled for transmission during the first TTI is overridden by the second set of data having the higher priority. The override indicator may be transmitted after the second set of data is transmitted. The one or more additional TTIs may be after the first TTI.

Abstract: Certain aspects of the present disclosure provide techniques for efficient bottleneck processing via dimensionality transformation. The techniques include receiving a tensor, and processing the tensor in a bottleneck block in a neural network model, comprising applying a space-to-depth tensor transformation, applying a depthwise convolution, and applying a depth-to-space tensor transformation.

Abstract: Certain aspects of the present disclosure provide techniques for generating fine depth maps for images of a scene based on semantic segmentation and segment-based refinement neural networks. An example method generally includes generating, through a segmentation neural network, a segmentation map based on an image of a scene. The segmentation map generally comprises a map segmenting the scene into a plurality of regions, and each region of the plurality of regions is generally associated with one of a plurality of categories. A first depth map of the scene is generated through a first depth neural network based on a depth measurement of the scene. A second depth map of the scene is generated through a depth refinement neural network based on the segmentation map and the first depth map. One or more actions are taken based on the second depth map of the scene.

Abstract: Systems and techniques are described herein for performing optical flow estimation for one or more frames. For example, a process can include determining an optical flow prediction associated with a plurality of frames. The process can include determining a position of at least one feature associated with a first frame and determining, based on the position of the at least one feature in the first frame and the optical flow prediction, a position estimate of a search area for searching for the at least one feature in a second frame.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for efficient processing of visual content using machine learning models. An example method generally includes generating, from an input, an embedding tensor for the input. The embedding tensor for the input is projected into a reduced-dimensional space projection of the embedding tensor based on a projection matrix. An attention value for the input is derived based on the reduced-dimensional space projection of the embedding tensor and a non-linear attention function. A match, in the reduced-dimensional space, is identified between a portion of the input and a corresponding portion of a target against which the input is evaluated based on the attention value for the input. One or more actions are taken based on identifying the match.

Abstract: A computer-implemented method includes receiving a first input. The first input is interpolated based on a first shift along a first dimension and a second shift along a second dimension. A first output is generated based on the interpolated first input. The first output corresponds to a vectorized bilinear shift of the first input for use in place of grid sampling algorithms.

Abstract: Embodiments described herein relate to a method, comprising: receiving input data at a convolutional neural network (CNN) model; generating a factorized computation network comprising a plurality of connections between a first layer of the CNN model and a second layer of the CNN model, wherein: the factorized computation network comprises N inputs, the factorized computation network comprises M outputs, and the factorized computation network comprises at least one path from every input of the N inputs to every output of the M outputs; setting a connection weight for a plurality of connections in the factorized computation network to 1 so that a weight density for the factorized computation network is <100%; performing fast pointwise convolution using the factorized computation network to generate fast pointwise convolution output; and providing the fast pointwise convolution output to the second layer of the CNN model.

Abstract: Systems and techniques are described herein for performing optical flow estimation for one or more frames. For example, a process can include determining an optical flow prediction associated with a plurality of frames. The process can include determining a position of at least one feature associated with a first frame and determining, based on the position of the at least one feature in the first frame and the optical flow prediction, a position estimate of a search area for searching for the at least one feature in a second frame. The process can include determining, from within the search area, a position of the at least one feature in the second frame.

Abstract: Aspects described herein provide a method including: receiving input data at a machine learning model, comprising: a plurality of processing layers; a plurality of gate logics; a plurality of gates; and a fully connected layer; determining based on a plurality of gate parameters associated with the plurality of gate logics, a subset of the plurality of processing layers with which to process the input data; processing the input data with the subset of the plurality of processing layers and the fully connected layer to generate an inference; determining a prediction loss based on the inference and a training label associated with the input data; determining an energy loss based on the subset of the plurality of processing layers used to process the input data; and optimizing the machine learning model based on: the prediction loss; the energy loss; and a prior probability associated with the training label.

Abstract: Certain aspects of the present disclosure provide techniques for domain adaptation. An input tensor comprising channel state information (CSI) for a wireless signal is determined, where each channel in the input tensor corresponds to a respective degree of freedom (DoF) in the wireless signal. A domain-adapted tensor is generated by processing the input tensor using a domain-adaptation network comprising, for each respective DoF in the wireless signal, a respective convolution path. The domain-adapted tensor is provided to a neural network trained for position estimation.

Abstract: Certain aspects of the present disclosure provide techniques for using shaped convolution kernels, comprising: receiving an input data patch, and processing the input data patch with a shaped kernel to generate convolution output.

Abstract: Certain aspects of the present disclosure provide techniques for parameterized activation functions. Input data is processed with at least one layer of the neural network model comprising a parameterized activation function, and at least one trainable parameter of the parameterized activation function is updated based at least in part on output from the at least one layer of the neural network model. The at least one trainable parameter may adjust at least one of a range over which the parameterized activation function is nonlinear or a shape of the parameterized activation function, and/or may adjust a location of at least one pivot of the parameterized activation function.

Abstract: Certain aspects of the present disclosure provide techniques for performing piecewise pointwise convolution, comprising: performing a first piecewise pointwise convolution on a first subset of data received via a first branch input at a piecewise pointwise convolution layer of a convolutional neural network (CNN) model; performing a second piecewise pointwise convolution on a second subset of data received via a second branch input at the piecewise pointwise convolution layer; determining a piecewise pointwise convolution output by summing a result of the first piecewise pointwise convolution and a result of the second piecewise pointwise convolution; and providing the piecewise pointwise convolution output to a second layer of the CNN model.

Abstract: Methods, systems, and devices for configuring a machine learning network are described. A device, which may be otherwise known as user equipment (UE), may support ultra-low power sensor applications. More specifically, the device may support memory block allocation of a machine learning network based on performance levels associated with the applications. For example, the device may identify a performance level associated with an application on the device. The device may determine that the performance level satisfies a condition, and subsequently determine a memory block allocation of a machine learning network of the device based on the performance level satisfying the condition. The memory block allocation may correspond to one or more connections of the machine learning network. Based on the memory block allocation, the device may adjust a quantity of memory blocks available for the machine learning network and process the application.

Abstract: Systems and techniques are described herein for performing optical flow estimation for one or more frames. For example, a process can include determining an optical flow prediction associated with a plurality of frames. The process can include determining a position of at least one feature associated with a first frame and determining, based on the position of the at least one feature in the first frame and the optical flow prediction, a position estimate of a search area for searching for the at least one feature in a second frame.

Abstract: Certain aspects of the present disclosure provide techniques for kernel expansion. An input data tensor is received at a first layer in a neural network, and a first convolution is performed for a first kernel, where the first kernel has a size greater than a preferred size. Performing the first convolution comprises generating a plurality of intermediate tensors by performing a plurality of intermediate convolutions using a plurality of intermediate kernels with a size of the preferred size, and accumulating the plurality of intermediate tensors to generate an output tensor for the first convolution.

Abstract: Disclosed are systems, methods, and non-transitory media for performing radio frequency sensing detection operations. For instance, radio frequency data can be received that is associated with at least one wireless device. The radio frequency data can be based on radio frequency signals reflected from a first object and received by the at least one wireless device. Training label data can also be obtained (e.g., from a labeling device, from the at least one wireless device, etc.). The training label data can be based at least in part on the first object and input data (e.g., received by the labeling device, the at least one wireless device, etc.). A sensing model can be generated based on the radio frequency data and the training label data.

Abstract: An apparatus may utilize an air interface to transmit and/or receive a transmission during a first TTI that includes a second set of data overriding a first set of data scheduled for transmission during the first TTI. The air interface may further be utilized to transmit and/or receive a transmission after a duration of time that is less than a total time duration of the first TTI that includes a control channel that is at least partially embedded in a data portion of a subframe. The control channel may include an override indicator configured to indicate that the first set of data scheduled for transmission during the first TTI is overridden by the second set of data having the higher priority. The override indicator may be transmitted after the second set of data is transmitted. The one or more additional TTIs may be after the first TTI.

Abstract: Certain aspects of the present disclosure provide a method for compressing an activation function, comprising: determining a plurality of difference values based on a difference between a target activation function and a reference activation function over a range of input values; determining a difference function based on the plurality of difference values; and performing an activation on input data using the reference activation function and a difference value based on the difference function.