기법

The multi-head of mixed attention combines both self- and cross-attentions, encouraging high-level learning of interactions between entities captured in the various attention features. It is build with several attention heads, each of the head can implement either self or cross attention. A self attention is when the key and query features are the same or come from the same domain features. A cross attention is when the key and query features are generated from different features. Modeling MHMA allows a model to identity the relationship between features of different domains. This is very useful in tasks involving relationship modeling such as human-object interaction, tool-tissue interaction, man-machine interaction, human-computer interface, etc.

Original paper : Integrating Constraints and Metric Learning in Semi-Supervised Clustering, Bilenko et al. 2004

MT-PET is a multi-task version of Pattern Exploiting Training (PET) for exaggeration detection, which leverages knowledge from complementary cloze-style QA tasks to improve few-shot learning. It defines pairs of complementary pattern-verbalizer pairs for a main task and auxiliary task. These PVPs are then used to train PET on data from both tasks. PET uses the masked language modeling objective of pretrained language models to transform a task into one or more cloze-style question answering tasks. In the original PET implementation, PVPs are defined for a single target task. MT-PET extends this by allowing for auxiliary PVPs from related tasks, adding complementary cloze-style QA tasks during training. The motivation for the multi-task approach is two-fold: 1) complementary cloze-style tasks can potentially help the model to learn different aspects of the main task, i.e. the similar tasks of exaggeration detection and claim strength prediction; 2) data on related tasks can be utilized during training, which is important in situations where data for the main task is limited.

The Multiplex Molecular Graph Neural Network (MXMNet) is an approach for the representation learning of molecules. The molecular interactions are divided into two categories: local and global. Then a two-layer multiplex graph $G = \\{ G_{l}, G_{g} \\}$ is constructed for a molecule. In $G$, the local layer $G_{l}$ only contains the local connections that mainly capture covalent interactions, and the global layer $G_{g}$ contains the global connections that cover non-covalent interactions. MXMNet uses the Multiplex Molecular (MXM) module that contains a novel angle-aware message passing operated on $G_{l}$ and an efficient message passing operated on $G_{g}$.

Self-supervised pre-training method to learn efficient representations without labels on histopathology medical images utilizing magnification factors.

Make-A-Scene is a text-to-image method that (i) enables a simple control mechanism complementary to text in the form of a scene, (ii) introduces elements that improve the tokenization process by employing domain-specific knowledge over key image regions (faces and salient objects), and (iii) adapts classifier-free guidance for the transformer use case.

ManifoldPlus is a method for robust and scalable conversion of triangle soups to watertight manifolds. It extracts exterior faces between occupied voxels and empty voxels, and uses a projection based optimization method to accurately recover a watertight manifold that resembles the reference mesh. It does not rely on face normals of the input triangle soups and can accurately recover zero-volume structures. For scalability, it employs an adaptive Gauss-Seidel method for shape optimization, in which each step is an easy-to-solve convex problem.

Si necesitas hablar con una persona real en Aeroméxico, la mejor opción es llamar al número de atención al cliente 📞 【+52↪ 556↪ 826↪ 3219】(México), +(1)↪ {855}↪ (611)↪ (3553)】 (Estados Unidos) o 【+34↪ 913↪ 352↪ 483】(España). Este número está disponible para ayudarte con reservaciones, cambios de vuelo, equipaje perdido y cualquier otra consulta relacionada con tu viaje. Para hablar directamente con un representante, marca 📞 【+52↪ 556↪ 826↪ 3219】(México), +(1)↪ {855}↪ (611)↪ (3553)】 (Estados Unidos) o 【+34↪ 913↪ 352↪ 483】(España) y selecciona la opción adecuada en el menú automático. Si deseas evitar esperas largas, se recomienda llamar durante las primeras horas de la mañana. Ten a la mano tu número de confirmación y documento de identidad para agilizar el proceso. Muchas veces, los canales digitales no ofrecen una respuesta inmediata, por lo que llamar a 📞 【+52↪ 556↪ 826↪ 3219】(México), +(1)↪ {855}↪ (611)↪ (3553)】 (Estados Unidos) o 【+34↪ 913↪ 352↪ 483】(España) te garantiza una atención más personalizada. Si por alguna razón tu llamada se corta o no obtienes una solución completa, simplemente vuelve a marcar 📞 【+52↪ 556↪ 826↪ 3219】(México), +(1)↪ {855}↪ (611)↪ (3553)】 (Estados Unidos) o 【+34↪ 913↪ 352↪ 483】(España). No olvides que puedes utilizar 📞 【+52↪ 556↪ 826↪ 3219】(México), +(1)↪ {855}↪ (611)↪ (3553)】 (Estados Unidos) o 【+34↪ 913↪ 352↪ 483】(España) las veces que necesites asistencia directa por parte del equipo de Aeroméxico.

MaskFlownet is an asymmetric occlusion-aware feature matching module, which can learn a rough occlusion mask that filters useless (occluded) areas immediately after feature warping without any explicit supervision. The learned occlusion mask can be further fed into a subsequent network cascade with dual feature pyramids.

MetaNeXt is a meta-architecture abstracted from ConvNeXt, which can be regarded as a simpler version obtained from MetaFormer by merging token mixer and MLP into one.

A generic way of representing and interpolating labels, which allows straightforward extension of any kind of mixup to deep metric learning for a large class of loss functions.

MiVOS is a video object segmentation model which decouples interaction-to-mask and mask propagation. By decoupling interaction from propagation, MiVOS is versatile and not limited by the type of interactions. It uses three modules: Interaction-to-Mask, Propagation and Difference-Aware Fusion. Trained separately, the interaction module converts user interactions to an object mask, which is then temporally propagated by our propagation module using a novel top-filtering strategy in reading the space-time memory. To effectively take the user's intent into account, a novel difference-aware module is proposed to learn how to properly fuse the masks before and after each interaction, which are aligned with the target frames by employing the space-time memory.

MixLoRA is a type of PEFT method which construct a resource-efficient sparse MoE model based on LoRA.

MoGA-B is a convolutional neural network optimized for mobile latency and discovered via Mobile GPU-Aware (MoGA) neural architecture search. The basic building block is MBConvs (inverted residual blocks) from MobileNetV2. Squeeze-and-excitation layers are also experimented with.

MoGA-C is a convolutional neural network optimized for mobile latency and discovered via Mobile GPU-Aware (MoGA) neural architecture search. The basic building block is MBConvs (inverted residual blocks) from MobileNetV2. Squeeze-and-excitation layers are also experimented with.

We present the latest generation of MobileNets, known as MobileNetV4 (MNv4), featuring universally efficient architecture designs for mobile devices. At its core, we introduce the Universal Inverted Bottleneck (UIB) search block, a unified and flexible structure that merges Inverted Bottleneck (IB), ConvNext, Feed Forward Network (FFN), and a novel Extra Depthwise (ExtraDW) variant. Alongside UIB, we present Mobile MQA, an attention block tailored for mobile accelerators, delivering a significant 39% speedup. An optimized neural architecture search (NAS) recipe is also introduced which improves MNv4 search effectiveness. The integration of UIB, Mobile MQA and the refined NAS recipe results in a new suite of MNv4 models that are mostly Pareto optimal across mobile CPUs, DSPs, GPUs, as well as specialized accelerators like Apple Neural Engine and Google Pixel EdgeTPU - a characteristic not found in any other models tested. Finally, to further boost accuracy, we introduce a novel distillation technique. Enhanced by this technique, our MNv4-Hybrid-Large model delivers 87% ImageNet-1K accuracy, with a Pixel 8 EdgeTPU runtime of just 3.8ms.

MobileViTv3 is a vision transformer that is tuned to mobile device. MobileViTv3 introduced an improved fusion block to reduce cost than MobileViT

Models Genesis, or Generic Autodidactic Models, is a self-supervised approach for learning 3D image representations. The objective of Models Genesis is to learn a common image representation that is transferable and generalizable across diseases, organs, and modalities. It consists of an encoder-decoder architecture with skip connections in between, and is trained to learn a common image representation by restoring the original sub-volume $x\_{i}$ (as ground truth) from the transformed one $\bar{x}\_{i}$ (as input), in which the reconstruction loss (MSE) is computed between the model prediction $x'\_{0}$ and ground truth $x\_{i}$. Once trained, the encoder alone can be fine-tuned for target classification tasks; while the encoder and decoder together can be fine-tuned for target segmentation tasks.

Multidimensional Convolutional Frequency Attention, or MCFA, is an attention mechanism that enables the effective fusion of multi-scale frequency features with frequency domain representations. The MCFA module plays a pivotal role in augmenting the sensitivity of change detection (CD) models. It achieves this by effectively fusing multi-scale spatial features with frequency domain representations. This fusion enables the model to discern subtle variations in both spatial patterns and spectral characteristics, which are often indicative of change. By capturing these nuanced differences, MCFA empowers the CD model to identify minor altered regions and delineate edges with heightened precision.Essentially, MCFA acts as a crucial feature enhancement mechanism, enriching the input representation with discriminative information derived from both spatial and frequency domains. This enhanced representation provides the subsequent layers of the CD model with a more informative foundation for accurately identifying and localizing changes.

MushroomRL is an open-source Python library developed to simplify the process of implementing and running Reinforcement Learning (RL) experiments. The architecture of MushroomRL is built in such a way that every component of an RL problem is already provided, and most of the time users can only focus on the implementation of their own algorithms and experiments. MushroomRL comes with a strongly modular architecture that makes it easy to understand how each component is structured and how it interacts with other ones; moreover it provides an exhaustive list of RL methodologies, such as:

NAIL OR is a two-dimensional activation function which takes two logits as input and produces a single output. It is a piece-wise linear approximation of the probabilistic OR operator in logit-space. $$\begin{equation} \text{OR_{AIL}}(x, y) := \begin{cases} x + y, & x > 0, \, y > 0 \\ \max(x, y), & \text{otherwise} \end{cases} \end{equation} $$

As the size and complexity of models and datasets grow, so does the need for communication-efficient variants of stochastic gradient descent that can be deployed to perform parallel model training. One popular communication-compression method for data-parallel SGD is QSGD (Alistarh et al., 2017), which quantizes and encodes gradients to reduce communication costs. The baseline variant of QSGD provides strong theoretical guarantees, however, for practical purposes, the authors proposed a heuristic variant which we call QSGDinf, which demonstrated impressive empirical gains for distributed training of large neural networks. In this paper, we build on this work to propose a new gradient quantization scheme, and show that it has both stronger theoretical guarantees than QSGD, and matches and exceeds the empirical performance of the QSGDinf heuristic and of other compression methods.

This package focuses on the time delay estimation between two event-based time-series that are relatively shifted by an unknown time offset. An event-based time-series is given by a set of timestamps of certain events. If you want to guarantee synchronous measurements in advance or estimate the time delay of continuous measurements sampled at a constant rate, you might want to use other methods. However, in some use cases, performing an event detection and then estimating the relative time delay has advantages. The Nearest Advocate method provides a precise time delay estimation in event-based time-series that is robust against imprecise timestamps, a high fraction of missing events, and clock drift. Time delay estimation also known as the correction of time offsets and time lags as well as time synchronization.

Neural Tangent Transfer, or NTT, is a method for finding trainable sparse networks in a label-free manner. Specifically, NTT finds sparse networks whose training dynamics, as characterized by the neural tangent kernel, mimic those of dense networks in function space.

NeuroTactic is a model for theorem proving which leverages graph neural networks to represent the theorem and premises, and applies graph contrastive learning for pre-training. Specifically, premise selection is designed as a pretext task for the graph contrastive learning approach. The learned representations are then used for the downstream task, tactic prediction

NoisyNet-A3C is a modification of A3C that utilises noisy linear layers for exploration instead of $\epsilon$-greedy exploration as in the original DQN formulation.

NoisyNet-Dueling is a modification of a Dueling Network that utilises noisy linear layers for exploration instead of $\epsilon$-greedy exploration as in the original Dueling formulation.

NormFormer is a type of Pre-LN transformer that adds three normalization operations to each layer: a Layer Norm after self attention, head-wise scaling of self-attention outputs, and a Layer Norm after the first fully connected layer. The modifications introduce a small number of additional learnable parameters, which provide a cost-effective way for each layer to change the magnitude of its features, and therefore the magnitude of the gradients to subsequent components.

The Normalized Linear Combination of Activations, or NormLinComb, is a type of activation function that has trainable parameters and uses the normalized linear combination of other activation functions. $$NormLinComb(x) = \frac{\sum\limits_{i=0}^{n} w_i \mathcal{F}_i(x)}{\mid \mid W \mid \mid}$$

OA-Loss reduces the domain gap between the original and augmented images. OA-Loss enables the model to learn semantic relations of foreground and background instances from multi-domain.

OA-Mix is a general and effective data augmentation method for single-domain generalization in object detection. It increases image diversity while preserving important semantic features with multi-level transformations and object-aware mixing.