Abstract: The present disclosure provides directed to new, more efficient neural network architectures. As one example, in some implementations, the neural network architectures of the present disclosure can include a linear bottleneck layer positioned structurally prior to and/or after one or more convolutional layers, such as, for example, one or more depthwise separable convolutional layers. As another example, in some implementations, the neural network architectures of the present disclosure can include one or more inverted residual blocks where the input and output of the inverted residual block are thin bottleneck layers, while an intermediate layer is an expanded representation. For example, the expanded representation can include one or more convolutional layers, such as, for example, one or more depthwise separable convolutional layers. A residual shortcut connection can exist between the thin bottleneck layers that play a role of an input and output of the inverted residual block.

Abstract: In one example, an LEP printer includes a belt rotatable in a loop, a series of multiple printing units along the belt each to apply an LEP ink color separation to the belt, a first heater to dry the color separations on the belt to a molten film at a drying temperature, a second heater to heat the molten film on the belt to a transfer temperature higher than the drying temperature, and a pressure roller near the belt to press a printable substrate against the belt at a nip between the pressure roller and the belt.

Abstract: The present disclosure provides directed to new, more efficient neural network architectures. As one example, in some implementations, the neural network architectures of the present disclosure can include a linear bottleneck layer positioned structurally prior to and/or after one or more convolutional layers, such as, for example, one or more depthwise separable convolutional layers. As another example, in some implementations, the neural network architectures of the present disclosure can include one or more inverted residual blocks where the input and output of the inverted residual block are thin bottleneck layers, while an intermediate layer is an expanded representation. For example, the expanded representation can include one or more convolutional layers, such as, for example, one or more depthwise separable convolutional layers. A residual shortcut connection can exist between the thin bottleneck layers that play a role of an input and output of the inverted residual block.

Abstract: According to an example, a mixing device comprises a first chamber including a solute addition member and a first mixer, a second chamber including a second mixer, a printing fluid pump, and a controller operatively connected to the solute addition member, the pump, the first mixer, and the second mixer. The controller is to control the pump to move printing fluid towards the printing fluid tank based on a printing fluid density level in the printing fluid tank, to control the solute addition member to add solids based on the printing fluid density level, and to control the first mixer and the second mixer to mix the added solids with the printing fluid moved by the pump.

Abstract: In an example of the disclosure, a heat exchange and flame arrest device for use with a subject gas includes tubing arranged to traverse a core, with the tubing defining a cooling fluid pathway. The core includes a gas flow inlet, a set of cooling fins, and a gas flow outlet. The gas flow inlet, the set of cooling fins, and the gas flow outlet collectively form a gas flow pathway for a subject gas. Each cooling fin of the set is positioned to form a gap between that cooling fin and an adjacent cooling fin. The gas flow inlet is to receive the subject gas. The combination of the gap between each cooling fin of the set and the cooling capacity of the cooling fluid pathway is sufficient to, if the subject gas has ignited, lower temperature of the subject gas to below autoignition temperature.

Abstract: The present disclosure provides systems and methods that enable parameter-efficient transfer learning, multi-task learning, and/or other forms of model re-purposing such as model personalization or domain adaptation. In particular, as one example, a computing system can obtain a machine-learned model that has been previously trained on a first training dataset to perform a first task. The machine-learned model can include a first set of learnable parameters. The computing system can modify the machine-learned model to include a model patch, where the model patch includes a second set of learnable parameters. The computing system can train the machine-learned model on a second training dataset to perform a second task that is different from the first task, which may include learning new values for the second set of learnable parameters included in the model patch while keeping at least some (e.g., all) of the first set of parameters fixed.

Abstract: In one example, an LEP printer includes a belt rotatable in a loop, a series of multiple printing units along the belt each to apply an LEP ink color separation to the belt, a first heater to dry the color separations on the belt to a molten film at a drying temperature, a second heater to heat the molten film on the belt to a transfer temperature higher than the drying temperature, and a pressure roller near the belt to press a printable substrate against the belt at a nip between the pressure roller and the belt.

Abstract: The present disclosure provides systems and methods that enable parameter-efficient transfer learning, multi-task learning, and/or other forms of model re-purposing such as model personalization or domain adaptation. In particular, as one example, a computing system can obtain a machine-learned model that has been previously trained on a first training dataset to perform a first task. The machine-learned model can include a first set of learnable parameters. The computing system can modify the machine-learned model to include a model patch, where the model patch includes a second set of learnable parameters. The computing system can train the machine-learned model on a second training dataset to perform a second task that is different from the first task, which may include learning new values for the second set of learnable parameters included in the model patch while keeping at least some (e.g., all) of the first set of parameters fixed.

Abstract: In one example, a system to clean imaging oil from a cleaning station in an LEP printer includes an electrode plate, an electrode belt rotatable in a loop past the electrode plate, a channel to carry unfiltered imaging oil from the cleaning station between the electrode plate and the electrode belt, and a voltage source operatively connected to the electrodes to establish an electric field that causes waste in the imaging oil between the electrodes to attach to the belt.

Abstract: A printing apparatus has a transfer member to receive an image and transfer the image to a substrate, and an airflow management unit disposed adjacent to the transfer member to define a channel between the transfer member and the airflow management unit. The airflow management unit has an injection mechanism to inject airflow into the channel, a suction mechanism to remove the airflow from the channel, and a vortex generator to generate a vortex within the channel.

Abstract: In an example of the disclosure, an intermediate transfer member cleaning system includes an intermediate transfer member (“blanket”), an endless cleaning surface, and a pyrolysis station. The blanket is to receive a thermoplastic print agent from a photoconductive element. The endless cleaning surface can be positioned to rotatably engage with the blanket to transfer a residue of the thermoplastic print agent from the blanket to the endless cleaning surface. The endless cleaning surface can be moved away from the blanket to enter a pyrolysis station. The pyrolysis station is for receiving the endless cleaning surface, for heating the endless cleaning surface to convert the residue to ash, and for removing the ash.

Abstract: In an example of the disclosure, a heat exchange and flame arrest device for use with a subject gas includes tubing arranged to traverse a core, with the tubing defining a cooling fluid pathway. The core includes a gas flow inlet, a set of cooling fins, and a gas flow outlet. The gas flow inlet, the set of cooling fins, and the gas flow outlet collectively form a gas flow pathway for a subject gas. Each cooling fin of the set is positioned to form a gap between that cooling fin and an adjacent cooling fin. The gas flow inlet is to receive the subject gas. The combination of the gap between each cooling fin of the set and the cooling capacity of the cooling fluid pathway is sufficient to, if the subject gas has ignited, lower temperature of the subject gas to below autoignition temperature.

Abstract: In an example, a heated air knife and a negative pressure component are controlled to heat a subject gas and direct the subject gas through an evaporation channel at a target gas velocity that is above a flame propagation velocity for the subject gas. The evaporation channel is defined in part by a first surface, a nozzle of the air knife, a mixing zone, and a second surface substantially opposite the first surface.

Abstract: The present disclosure provides directed to new, more efficient neural network architectures. As one example, in some implementations, the neural network architectures of the present disclosure can include a linear bottleneck layer positioned structurally prior to and/or after one or more convolutional layers, such as, for example, one or more depthwise separable convolutional layers. As another example, in some implementations, the neural network architectures of the present disclosure can include one or more inverted residual blocks where the input and output of the inverted residual block are thin bottleneck layers, while an intermediate layer is an expanded representation. For example, the expanded representation can include one or more convolutional layers, such as, for example, one or more depthwise separable convolutional layers. A residual shortcut connection can exist between the thin bottleneck layers that play a role of an input and output of the inverted residual block.

Abstract: In an example of the disclosure, an intermediate transfer member cleaning system includes an intermediate transfer member (“blanket”), an endless cleaning surface, and a pyrolysis station. The blanket is to receive a thermoplastic print agent from a photoconductive element. The endless cleaning surface can be positioned to rotatably engage with the blanket to transfer a residue of the thermoplastic print agent from the blanket to the endless cleaning surface. The endless cleaning surface can be moved away from the blanket to enter a pyrolysis station. The pyrolysis station is for receiving the endless cleaning surface, for heating the endless cleaning surface to convert the residue to ash, and for removing the ash.

Abstract: In one example, an exhaust hood includes an enclosure having a perimeter defining an exhaust area and a fluid flow path. The fluid flow path includes a perimeter intake slot through which air is sucked into the flow path during an exhaust operation and a flow channel in fluid communication with the perimeter intake slot and configured to carry fluid away from the intake slot and out of the enclosure.

Abstract: The present disclosure provides directed to new, more efficient neural network architectures. As one example, in some implementations, the neural network architectures of the present disclosure can include a linear bottleneck layer positioned structurally prior to and/or after one or more convolutional layers, such as, for example, one or more depthwise separable convolutional layers. As another example, in some implementations, the neural network architectures of the present disclosure can include one or more inverted residual blocks where the input and output of the inverted residual block are thin bottleneck layers, while an intermediate layer is an expanded representation. For example, the expanded representation can include one or more convolutional layers, such as, for example, one or more depthwise separable convolutional layers. A residual shortcut connection can exist between the thin bottleneck layers that play a role of an input and output of the inverted residual block.

Abstract: A printing apparatus has a transfer member to receive an image and transfer the image to a substrate, and an airflow management unit disposed adjacent to the transfer member to define a channel between the transfer member and the airflow management unit. The airflow management unit has an injection mechanism to inject airflow into the channel, a suction mechanism to remove the airflow from the channel, and a vortex generator to generate a vortex within the channel.

Abstract: Described herein is a liquid electrophotographic photovoltaic ink composition comprising: a dispersion of a material with a perovskite structure, a thermoplastic resin and conductive particles in a carrier liquid; wherein the material with a perovskite structure has a chemical formula selected from ABX3 and A2BX6; wherein A is a cation, B is a cation and X is an anion; and wherein the thermoplastic resin comprises: a copolymer of an alkylene monomer and a monomer having acidic side groups; and/or a copolymer of an alkylene monomer and an ethylenically unsaturated monomer comprising an epoxide; and/or a copolymer of an alkylene monomer, an ethylenically unsaturated monomer comprising an epoxide, and a monomer selected from a monomer having acidic side groups, a monomer having ester side groups and a mixture thereof. Also described is a method of producing a photovoltaic cell using the LEP ink and the printed cell produced by the method.

Abstract: The present disclosure provides a new type of generalized artificial neural network where neurons and synapses maintain multiple states. While classical gradient-based backpropagation in artificial neural networks can be seen as a special case of a two-state network where one state is used for activations and another for gradients with update rules derived from the chain rule, example implementations of the generalized framework proposed herein may additionally: have neither explicit notion of nor ever receive gradients; contain more than two states; and/or implement or apply learned (e.g., meta-learned) update rules that control updates to the state(s) of the neuron during forward and/or backward propagation of information.