기법

Siren, or Sinusoidal Representation Network, is a periodic activation function for implicit neural representations. Specifically it uses the sine as a periodic activation function: $$ \Phi\left(x\right) = \textbf{W}\_{n}\left(\phi\_{n-1} \circ \phi\_{n-2} \circ \dots \circ \phi\_{0} \right) $$

Spatial Gating Unit, or SGU, is a gating unit used in the gMLP architecture to captures spatial interactions. To enable cross-token interactions, it is necessary for the layer $s(\cdot)$ to contain a contraction operation over the spatial dimension. The layer $s(\cdot)$ is formulated as the output of linear gating: $$ s(Z)=Z \odot f\_{W, b}(Z) $$ where $\odot$ denotes element-wise multiplication. For training stability, the authors find it critical to initialize $W$ as near-zero values and $b$ as ones, meaning that $f\_{W, b}(Z) \approx 1$ and therefore $s(Z) \approx Z$ at the beginning of training. This initialization ensures each gMLP block behaves like a regular FFN at the early stage of training, where each token is processed independently, and only gradually injects spatial information across tokens during the course of learning. The authors find it further effective to split $Z$ into two independent parts $\left(Z\_{1}, Z\_{2}\right)$ along the channel dimension for the gating function and for the multiplicative bypass: $$ s(Z)=Z\_{1} \odot f\_{W, b}\left(Z\_{2}\right) $$ They also normalize the input to $f\_{W, b}$ which empirically improved the stability of large NLP models.

TabTransformer is a deep tabular data modeling architecture for supervised and semi-supervised learning. The TabTransformer is built upon self-attention based Transformers. The Transformer layers transform the embeddings of categorical features into robust contextual embeddings to achieve higher prediction accuracy. As an overview, the architecture comprises a column embedding layer, a stack of $N$ Transformer layers, and a multi-layer perceptron (MLP). The contextual embeddings (outputted by the Transformer layer) are concatenated along with continuous features which is inputted to an MLP. The loss function is then minimized to learn all the parameters in an end-to-end learning.

Zero Redundancy Optimizer (ZeRO) is a sharded data parallel method for distributed training. ZeRODP removes the memory state redundancies across data-parallel processes by partitioning the model states instead of replicating them, and it retains the compute/communication efficiency by retaining the computational granularity and communication volume of DP using a dynamic communication schedule during training.

AGCN is a novel spectral graph convolution network that feed on original data of diverse graph structures. Image credit: Adaptive Graph Convolutional Neural Networks

Context Enhancement Module (CEM) is a feature extraction module used in object detection (specifically, ThunderNet) which aims to to enlarge the receptive field. The key idea of CEM is to aggregate multi-scale local context information and global context information to generate more discriminative features. In CEM, the feature maps from three scales are merged: $C\_{4}$, $C\_{5}$ and $C\_{glb}$. $C\_{glb}$ is the global context feature vector by applying a global average pooling on $C\_{5}$. We then apply a 1 × 1 convolution on each feature map to squeeze the number of channels to $\alpha \times p \times p = 245$. Afterwards, $C\_{5}$ is upsampled by 2× and $C\_{glb}$ is broadcast so that the spatial dimensions of the three feature maps are equal. At last, the three generated feature maps are aggregated. By leveraging both local and global context, CEM effectively enlarges the receptive field and refines the representation ability of the thin feature map. Compared with prior FPN structures, CEM involves only two 1×1 convolutions and a fc layer.

Decentralized Distributed Proximal Policy Optimization (DD-PPO) is a method for distributed reinforcement learning in resource-intensive simulated environments. DD-PPO is distributed (uses multiple machines), decentralized (lacks a centralized server), and synchronous (no computation is ever `stale'), making it conceptually simple and easy to implement. Proximal Policy Optimization, or PPO, is a policy gradient method for reinforcement learning. The motivation was to have an algorithm with the data efficiency and reliable performance of TRPO, while using only first-order optimization. Let $r\_{t}\left(\theta\right)$ denote the probability ratio $r\_{t}\left(\theta\right) = \frac{\pi\_{\theta}\left(a\_{t}\mid{s\_{t}}\right)}{\pi\_{\theta\_{old}}\left(a\_{t}\mid{s\_{t}}\right)}$, so $r\left(\theta\_{old}\right) = 1$. TRPO maximizes a “surrogate” objective: $$ L^{v}\left({\theta}\right) = \hat{\mathbb{E}}\_{t}\left[\frac{\pi\_{\theta}\left(a\_{t}\mid{s\_{t}}\right)}{\pi\_{\theta\_{old}}\left(a\_{t}\mid{s\_{t}}\right)})\hat{A}\_{t}\right] = \hat{\mathbb{E}}\_{t}\left[r\_{t}\left(\theta\right)\hat{A}\_{t}\right] $$ As a general abstraction, DD-PPO implements the following: at step $k$, worker $n$ has a copy of the parameters, $\theta^k_n$, calculates the gradient, $\delta \theta^k_n$, and updates $\theta$ via $$ \theta^{k+1}\_n = \text{ParamUpdate}\Big(\theta^{k}\_n, \text{AllReduce}\big(\delta \theta^k\_1, \ldots, \delta \theta^k\_N\big)\Big) = \text{ParamUpdate}\Big(\theta^{k}\_n, \frac{1}{N} \sum_{i=1}^{N} { \delta \theta^k_i} \Big) $$ where $\text{ParamUpdate}$ is any first-order optimization technique (e.g. gradient descent) and $\text{AllReduce}$ performs a reduction (e.g. mean) over all copies of a variable and returns the result to all workers. Distributed DataParallel scales very well (near-linear scaling up to 32,000 GPUs), and is reasonably simple to implement (all workers synchronously running identical code).

Denoised Smoothing is a method for obtaining a provably robust classifier from a fixed pretrained one, without any additional training or fine-tuning of the latter. The basic idea is to prepend a custom-trained denoiser before the pretrained classifier, and then apply randomized smoothing. Randomized smoothing is a certified defense that converts any given classifier $f$ into a new smoothed classifier $g$ that is characterized by a non-linear Lipschitz property. When queried at a point $x$, the smoothed classifier $g$ outputs the class that is most likely to be returned by $f$ under isotropic Gaussian perturbations of its inputs. Unfortunately, randomized smoothing requires that the underlying classifier $f$ is robust to relatively large random Gaussian perturbations of the input, which is not the case for off-the-shelf pretrained models. By applying our custom-trained denoiser to the classifier $f$, we can effectively make $f$ robust to such Gaussian perturbations, thereby making it “suitable” for randomized smoothing.

Embedded Gaussian Affinity is a type of affinity or self-similarity function between two points $\mathbf{x\_{i}}$ and $\mathbf{x\_{j}}$ that uses a Gaussian function in an embedding space: $$ f\left(\mathbf{x\_{i}}, \mathbf{x\_{j}}\right) = e^{\theta\left(\mathbf{x\_{i}}\right)^{T}\phi\left(\mathbf{x\_{j}}\right)} $$ Here $\theta\left(x\_{i}\right) = W\_{θ}x\_{i}$ and $\phi\left(x\_{j}\right) = W\_{φ}x\_{j}$ are two embeddings. Note that the self-attention module used in the original Transformer model is a special case of non-local operations in the embedded Gaussian version. This can be seen from the fact that for a given $i$, $\frac{1}{\mathcal{C}\left(\mathbf{x}\right)}\sum\_{\forall{j}}f\left(\mathbf{x}\_{i}, \mathbf{x}\_{j}\right)g\left(\mathbf{x}\_{j}\right)$ becomes the softmax computation along the dimension $j$. So we have $\mathbf{y} = \text{softmax}\left(\mathbf{x}^{T}W^{T}\_{\theta}W\_{\phi}\mathbf{x}\right)g\left(\mathbf{x}\right)$, which is the self-attention form in the Transformer model. This shows how we can relate this recent self-attention model to the classic computer vision method of non-local means.

GCT first collects global information by computing the l2-norm of each channel. Next, a learnable vector $ \alpha $ is applied to scale the feature. Then a competition mechanism is adopted by channel normalization to interact between channels. Unlike previous methods, GCT first collects global information by computing the $l_{2}$-norm of each channel. Next, a learnable vector $\alpha$ is applied to scale the feature. Then a competition mechanism is adopted by channel normalization to interact between channels. Like other common normalization methods, a learnable scale parameter $\gamma$ and bias $\beta$ are applied to rescale the normalization. However, unlike previous methods, GCT adopts tanh activation to control the attention vector. Finally, it not only multiplies the input by the attention vector but also adds an identity connection. GCT can be written as: \begin{align} s = F_\text{gct}(X, \theta) & = \tanh (\gamma CN(\alpha \text{Norm}(X)) + \beta) \end{align} \begin{align} Y & = s X + X \end{align} where $\alpha$, $\beta$ and $\gamma$ are trainable parameters. $\text{Norm}(\cdot)$ indicates the $L2$-norm of each channel. $CN$ is channel normalization. A GCT block has fewer parameters than an SE block, and as it is lightweight, can be added after each convolutional layer of a CNN.

A Gated Linear Network, or GLN, is a type of backpropagation-free neural architecture. What distinguishes GLNs from contemporary neural networks is the distributed and local nature of their credit assignment mechanism; each neuron directly predicts the target, forgoing the ability to learn feature representations in favor of rapid online learning. Individual neurons can model nonlinear functions via the use of data-dependent gating in conjunction with online convex optimization. GLNs are feedforward networks composed of many layers of gated geometric mixing neurons as shown in the Figure . Each neuron in a given layer outputs a gated geometric mixture of the predictions from the previous layer, with the final layer consisting of just a single neuron. In a supervised learning setting, a $\mathrm{GLN}$ is trained on (side information, base predictions, label) triplets $\left(z\_{t}, p\_{t}, x\_{t}\right)_{t=1,2,3, \ldots}$ derived from input-label pairs $\left(z\_{t}, x\_{t}\right)$. There are two types of input to neurons in the network: the first is the side information $z\_{t}$, which can be thought of as the input features; the second is the input to the neuron, which will be the predictions output by the previous layer, or in the case of layer 0 , some (optionally) provided base predictions $p\_{t}$ that typically will be a function of $z\_{t} .$ Each neuron will also take in a constant bias prediction, which helps empirically and is essential for universality guarantees. Weights are learnt in a Gated Linear Network using Online Gradient Descent (OGD) locally at each neuron. They key observation is that as each neuron $(i, k)$ in layers $i>0$ is itself a gated geometric mixture, all of these neurons can be thought of as individually predicting the target. Given side information $z$ , each neuron $(i, k)$ suffers a loss convex in its active weights $u:=w\_{i k c\_{i k}(z)}$ of $$ \ell\_{t}(u):=-\log \left(\operatorname{GEO}\_{u}\left(x_{t} ; p\_{i-1}\right)\right) $$

Gradual self-training is a method for semi-supervised domain adaptation. The goal is to adapt an initial classifier trained on a source domain given only unlabeled data that shifts gradually in distribution towards a target domain. This comes up for example in applications ranging from sensor networks and self-driving car perception modules to brain-machine interfaces, where machine learning systems must adapt to data distributions that evolve over time. The gradual self-training algorithm begins with a classifier $w_0$ trained on labeled examples from the source domain (Figure a). For each successive domain $P_t$, the algorithm generates pseudolabels for unlabeled examples from that domain, and then trains a regularized supervised classifier on the pseudolabeled examples. The intuition, visualized in the Figure, is that after a single gradual shift, most examples are pseudolabeled correctly so self-training learns a good classifier on the shifted data, but the shift from the source to the target can be too large for self-training to correct.

In the context of image enhancement, maximizing NIMA score as a prior can increase the likelihood of enhancing perceptual quality of an image.

Some object detection datasets contain an overwhelming number of easy examples and a small number of hard examples. Automatic selection of these hard examples can make training more effective and efficient. OHEM, or Online Hard Example Mining, is a bootstrapping technique that modifies SGD to sample from examples in a non-uniform way depending on the current loss of each example under consideration. The method takes advantage of detection-specific problem structure in which each SGD mini-batch consists of only one or two images, but thousands of candidate examples. The candidate examples are subsampled according to a distribution that favors diverse, high loss instances.

Pretext-Invariant Representation Learning (PIRL, pronounced as “pearl”) learns invariant representations based on pretext tasks. PIRL is used with a commonly used pretext task that involves solving jigsaw puzzles. Specifically, PIRL constructs image representations that are similar to the representation of transformed versions of the same image and different from the representations of other images.

Positional Encoding Generator, or PEG, is a module used in the Conditional Position Encoding position embeddings. It dynamically produce the positional encodings conditioned on the local neighborhood of an input token. To condition on the local neighbors, we first reshape the flattened input sequence $X \in \mathbb{R}^{B \times N \times C}$ of DeiT back to $X^{\prime} \in \mathbb{R}^{B \times H \times W \times C}$ in the 2 -D image space. Then, a function (denoted by $\mathcal{F}$ in the Figure) is repeatedly applied to the local patch in $X^{\prime}$ to produce the conditional positional encodings $E^{B \times H \times W \times C} .$ PEG can be efficiently implemented with a 2-D convolution with kernel $k(k \geq 3)$ and $\frac{k-1}{2}$ zero paddings. Note that the zero paddings here are important to make the model be aware of the absolute positions, and $\mathcal{F}$ can be of various forms such as separable convolutions and many others.

A Pyramidal Bottleneck Residual Unit is a type of residual unit where the number of channels gradually increases as a function of the depth at which the layer occurs, which is similar to a pyramid structure of which the shape gradually widens from the top downwards. It also consists of a bottleneck using 1x1 convolutions. It was introduced as part of the PyramidNet architecture.

A Pyramidal Residual Unit is a type of residual unit where the number of channels gradually increases as a function of the depth at which the layer occurs, which is similar to a pyramid structure of which the shape gradually widens from the top downwards. It was introduced as part of the PyramidNet architecture.

Receptive Field Block (RFB) is a module for strengthening the deep features learned from lightweight CNN models so that they can contribute to fast and accurate detectors. Specifically, RFB makes use of multi-branch pooling with varying kernels corresponding to RFs of different sizes, applies dilated convolution layers to control their eccentricities, and reshapes them to generate final representation.

Rotary Position Embedding, or RoPE, is a type of position embedding which encodes absolute positional information with rotation matrix and naturally incorporates explicit relative position dependency in self-attention formulation. Notably, RoPE comes with valuable properties such as flexibility of being expand to any sequence lengths, decaying inter-token dependency with increasing relative distances, and capability of equipping the linear self-attention with relative position encoding.

STAC is a semi-supervised framework for visual object detection along with a data augmentation strategy. STAC deploys highly confident pseudo labels of localized objects from an unlabeled image and updates the model by enforcing consistency via strong augmentations. We generate pseudo labels (i.e., bounding boxes and their class labels) for unlabeled data using test-time inference, including NMS , of the teacher model trained with labeled data. We then compute unsupervised loss with respect to pseudo labels whose confidence scores are above a threshold $\tau$ . The strong augmentations are applied for augmentation consistency during the model training. Target boxes are augmented when global geometric transformations are used.

The overhead cost of training multiple deep neural networks could be very high in terms of the training time, hardware, and computational resource requirement and often acts as obstacle for creating deep ensembles. To overcome these barriers Huang et al. proposed a unique method to create ensemble which at the cost of training one model, yields multiple constituent model snapshots that can be ensembled together to create a strong learner. The core idea behind the concept is to make the model converge to several local minima along the optimization path and save the model parameters at these local minima points. During the training phase, a neural network would traverse through many such points. The lowest of all such local minima is known as the Global Minima. The larger the model, more are the number of parameters and larger the number of local minima points. This implies, there are discrete sets of weights and biases, at which the model is making fewer errors. So, every such minimum can be considered a weak but a potential learner model for the problem being solved. Multiple such snapshot of weights and biases are recorded which can later be ensembled to get a better generalized model which makes the least amount of mistakes.

Spatial Feature Transform, or SFT, is a layer that generates affine transformation parameters for spatial-wise feature modulation, and was originally proposed within the context of image super-resolution. A Spatial Feature Transform (SFT) layer learns a mapping function $\mathcal{M}$ that outputs a modulation parameter pair $(\mathbf{\gamma}, \mathbf{\beta})$ based on some prior condition $\Psi$. The learned parameter pair adaptively influences the outputs by applying an affine transformation spatially to each intermediate feature maps in an SR network. During testing, only a single forward pass is needed to generate the HR image given the LR input and segmentation probability maps. More precisely, the prior $\Psi$ is modeled by a pair of affine transformation parameters $(\mathbf{\gamma}, \mathbf{\beta})$ through a mapping function $\mathcal{M}: \Psi \mapsto(\mathbf{\gamma}, \mathbf{\beta})$. Consequently, $$ \hat{\mathbf{y}}=G_{\mathbf{\theta}}(\mathbf{x} \mid \mathbf{\gamma}, \mathbf{\beta}), \quad(\mathbf{\gamma}, \mathbf{\beta})=\mathcal{M}(\Psi) $$ After obtaining $(\mathbf{\gamma}, \mathbf{\beta})$ from conditions, the transformation is carried out by scaling and shifting feature maps of a specific layer: $$ \operatorname{SFT}(\mathbf{F} \mid \mathbf{\gamma}, \mathbf{\beta})=\mathbf{\gamma} \odot \mathbf{F}+\mathbf{\beta} $$ where $\mathbf{F}$ denotes the feature maps, whose dimension is the same as $\gamma$ and $\mathbf{\beta}$, and $\odot$ is referred to element-wise multiplication, i.e., Hadamard product. Since the spatial dimensions are preserved, the SFT layer not only performs feature-wise manipulation but also spatial-wise transformation.

TDN, or Temporaral Difference Network, is an action recognition model that aims to capture multi-scale temporal information. To fully capture temporal information over the entire video, the TDN is established with a two-level difference modeling paradigm. Specifically, for local motion modeling, temporal difference over consecutive frames is used to supply 2D CNNs with finer motion pattern, while for global motion modeling, temporal difference across segments is incorporated to capture long-range structure for motion feature excitation.

UL2 is a unified framework for pretraining models that are universally effective across datasets and setups. UL2 uses Mixture-of-Denoisers (MoD), a pre-training objective that combines diverse pre-training paradigms together. UL2 introduces a notion of mode switching, wherein downstream fine-tuning is associated with specific pre-training schemes.

VarifocalNet is a method aimed at accurately ranking a huge number of candidate detections in object detection. It consists of a new loss function, named Varifocal Loss, for training a dense object detector to predict the IACS, and a new efficient star-shaped bounding box feature representation for estimating the IACS and refining coarse bounding boxes. Combining these two new components and a bounding box refinement branch, results in a dense object detector on the FCOS architecture, what the authors call VarifocalNet or VFNet for short.

Varifocal Loss is a loss function for training a dense object detector to predict the IACS, inspired by focal loss. Unlike the focal loss that deals with positives and negatives equally, Varifocal Loss treats them asymmetrically. $$ VFL\left(p, q\right) = −q\left(q\log\left(p\right) + \left(1 − q\right)\log\left(1 − p\right)\right) \text{ if } q > 0 $$ $$ VFL\left(p, q\right) = −\alpha{p^{\gamma}}\log\left(1-p\right) $$ where $p$ is the predicted IACS and $q$ is the target IoU score. For a positive training example, $q$ is set as the IoU between the generated bounding box and the ground-truth one (gt IoU), whereas for a negative training example, the training target $q$ for all classes is $0$.

BLANC is an automatic estimation approach for document summary quality. The goal is to measure the functional performance of a summary with an objective, reproducible, and fully automated method. BLANC achieves this by measuring the performance boost gained by a pre-trained language model with access to a document summary while carrying out its language understanding task on the document's text.

Bridge-net is an audio model block used in the ClariNet text-to-speech architecture. Bridge-net maps frame-level hidden representation to sample-level through several convolution blocks and transposed convolution layers interleaved with softsign non-linearities.

ClariNet is an end-to-end text-to-speech architecture. Unlike previous TTS systems which use text-to-spectogram models with a separate waveform synthesizer (vocoder), ClariNet is a text-to-wave architecture that is fully convolutional and can be trained from scratch. In ClariNet, the WaveNet module is conditioned on the hidden states instead of the mel-spectogram. The architecture is otherwise based on Deep Voice 3.

Conditional Positional Encoding, or CPE, is a type of positional encoding for vision transformers. Unlike previous fixed or learnable positional encodings, which are predefined and independent of input tokens, CPE is dynamically generated and conditioned on the local neighborhood of the input tokens. As a result, CPE aims to generalize to the input sequences that are longer than what the model has ever seen during training. CPE can also keep the desired translation-invariance in the image classification task. CPE can be implemented with a Position Encoding Generator (PEG) and incorporated into the current Transformer framework.

CornerNet is an object detection model that detects an object bounding box as a pair of keypoints, the top-left corner and the bottom-right corner, using a single convolution neural network. By detecting objects as paired keypoints, we eliminate the need for designing a set of anchor boxes commonly used in prior single-stage detectors. It also utilises corner pooling, a new type of pooling layer than helps the network better localize corners.

In stochastic processes, chaos theory and time series analysis, detrended fluctuation analysis (DFA) is a method for determining the statistical self-affinity of a signal. It is useful for analysing time series that appear to be long-memory processes (diverging correlation time, e.g. power-law decaying autocorrelation function) or 1/f noise. The obtained exponent is similar to the Hurst exponent, except that DFA may also be applied to signals whose underlying statistics (such as mean and variance) or dynamics are non-stationary (changing with time). It is related to measures based upon spectral techniques such as autocorrelation and Fourier transform.

DiffPool is a differentiable graph pooling module that can generate hierarchical representations of graphs and can be combined with various graph neural network architectures in an end-to-end fashion. DiffPool learns a differentiable soft cluster assignment for nodes at each layer of a deep GNN, mapping nodes to a set of clusters, which then form the coarsened input for the next GNN layer. Description and image from: Hierarchical Graph Representation Learning with Differentiable Pooling