Machine Learning: A Form Of Applied Statistics

Machine Learning

Machine learning is essentially a form of applied statistics with increased emphasis on the use of computers to statistically estimate complicated functions and a decreased emphasis on proving confidence intervals around these functions.

Machine learning algorithms have settings called hyperparameters that must be determined external to the learning algorithm itself.

1. Learning Algorithms

A machine learning algorithm is an algorithm that is able to learn from data. But what do we mean by learning? Mitchell (1997) provides the definition “A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.”

1) The Task, T

Machine learning allows us to tackle tasks that are too difficult to solve with fixed programs written and designed by human beings. From a scientific and philosophical point of view, machine learning is interesting because developing our understanding of machine learning entails developing our understanding of the principles that underlie intelligence.

In this relatively formal definition of the word “task,” the process of learning itself is not the task.

Machine learning tasks are usually described in terms of how the machine learning system should process an example. An example is a collection of features that have been quantitatively measured from some object or event that we want the machine learning system to process.

Many kinds of tasks can be solved with machine learning. Some of the most common machine learning tasks include the following:

• Classification: In this type of task, the computer program is asked to specify which of k categories some input belongs to. To solve this task, the learning algorithm is usually asked to produce a function f :→ {1, . . . , k}. When y = f(x), the model assigns an input described by vector x to a category identified by numeric code y. There are other variants of the classification task, for example, where f outputs a probability distribution over classes. An example of a classification task is object recognition, where the input is an image (usually described as a set of pixel brightness values), and the output is a numeric code identifying the object in the image.
• Classification with missing inputs: Classification becomes more challenging if the computer program is not guaranteed that every measurement in its input vector will always be provided. In order to solve the classification task, the learning algorithm only has to define a single function mapping from a vector input to a categorical output.
• Regression: In this type of task, the computer program is asked to predict a numerical value given some input. To solve this task, the learning algorithm is asked to output a function f: → . This type of task is similar to classification, except that the format of output is different. An example of a regression task is the prediction of future prices of securities.
• Transcription: In this type of task, the machine learning system is asked to observe a relatively unstructured representation of some kind of data and transcribe it into discrete, textual form. Google Street View uses deep learning to process address numbers in this way (Goodfellow et al., 2014d). Another example is speech recognition, where the computer program is provided an audio waveform and emits a sequence of characters or word ID codes describing the words that were spoken in the audio recording. Deep learning is a crucial component of modern speech recognition systems used at major companies including Microsoft, IBM and Google (Hinton et al.,2012b)

2) The Performance Measure, P

Machine learning algorithms can be broadly categorized as unsupervised or supervised by what kind of experience they are allowed to have during the learning process.

In order to evaluate the abilities of a machine learning algorithm, we must design a quantitative measure of its performance. Usually, this performance measure P is specific to the task T being carried out by the system.

For tasks such as classification, classification with missing inputs, and transcription, we often measure the accuracy of the model. Accuracy is just the proportion of examples for which the model produces the correct output.

3) The Experience, E

Most of the learning algorithms can be understood as being allowed to experience an entire dataset. A dataset is a collection of many examples.

Unsupervised learning algorithms experience a dataset containing many features, then learn useful properties of the structure of this dataset. In the context of deep learning, we usually want to learn the entire probability distribution that generated a dataset, whether explicitly as in density estimation or implicitly for tasks like synthesis or denoising. Some other unsupervised learning algorithms perform other roles, like clustering, which consists of dividing the dataset into clusters of similar examples.

Supervised learning algorithms experience a dataset containing features, but each example is also associated with a label or target. For example, the Iris dataset is annotated with the species of each iris plant. A supervised learning algorithm can study the Iris dataset and learn to classify iris plants into three different species based on their measurements.

Convolutional Networks

Convolutional networks (LeCun, 1989), also known as convolutional neural networks or CNNs, are a specialized kind of neural network for processing data that has a known, grid-like topology.

The name “convolutional neural network” indicates that the network employs a mathematical operation called convolution. Convolution is a specialized kind of linear operation. Convolutional networks are simply neural networks that use convolution in place of general matrix multiplication in at least one of their layers.

Object detection

Object detection is a computer vision technique that allows us to identify and locate objects in an image or video. With this kind of identification and localization, object detection can be used to count objects in a scene and determine and track their precise locations, all while accurately labeling them.

Object detection is commonly confused with image recognition, so before we proceed, it’s important that we clarify the distinctions between them.

Image recognition assigns a label to an image. A picture of a dog receives the label “dog”. A picture of two dogs, still receives the label “dog”. Object detection, on the other hand, draws a box around each dog and labels the box “dog”. The model predicts where each object is and what label should be applied. In that way, object detection provides more information about an image than recognition.

Broadly speaking, object detection can be broken down into:

• machine learning-based approaches
• deep learning-based approaches.

Modes and types of object detectionIn more traditional ML-based approaches, computer vision techniques are used to look at various features of an image, such as the color histogram or edges, to identify groups of pixels that may belong to an object. These features are then fed into a regression model that predicts the location of the object along with its label.

On the other hand, deep learning-based approaches employ convolutional neural networks (CNNs) to perform end-to-end, unsupervised object detection, in which features don’t need to be defined and extracted separately.

Why is object detection important?

Object detection is inextricably linked to other similar computer vision techniques like image recognition and image segmentation, in that it helps us understand and analyze scenes in images or video.

But there are important differences. Image recognition only outputs a class label for an identified object, and image segmentation creates a pixel-level understanding of a scene’s elements. What separates object detection from these other tasks is its unique ability to locate objects within an image or video. This then allows us to count and then track those objects.

It is a core requirement for any self-driving car perception stack. Object detection is a top level module required to identify the locations of vehicles, pedestrians, signs, and lights, so our car knows where and how it can drive.

How does object detection work?

Now that we know a bit about what object detection is, the distinctions between different types of object detection, and what it can be used for, let’s explore in more depth how it actually works.

we’ll look at several deep learning-based approaches to object detection and assess their advantages and limitations. Just as a reminder—for the purposes of this overview, we’re going to look at the approaches that use neural networks, which have become the state-of-the-art methods for object detection. In this section, we’ll look at several deep learning-based approaches to object detection and assess their advantages and limitations.

Basic structure

Deep learning-based object detection models typically have two parts. An encoder takes an image as input and runs it through a series of blocks and layers that learn to extract statistical features used to locate and label objects. Outputs from the encoder are then passed to a decoder, which predicts bounding boxes and labels for each object.

The simplest decoder is a pure regressor. The regressor is connected to the output of the encoder and predicts the location and size of each bounding box directly. The output of the model is the X, Y coordinate pair for the object and its extent in the image. Though simple, this type of model is limited. You need to specify the number of boxes ahead of time. If your image has two dogs, but your model was only designed to detect a single object, one will go unlabeled. However, if you know the number of objects you need to predict in each image ahead of time, pure regressor-based models may be a good option.

An extension of the regressor approach is a region proposal network. In this decoder, the model proposes regions of an image where it believes an object might reside. The pixels belonging to these regions are then fed into a classification subnetwork to determine a label (or reject the proposal). It then runs the pixels containing those regions through a classification network. The benefit of this method is a more accurate, flexible model that can propose arbitrary numbers of regions that may contain a bounding box. The added accuracy, though, comes at the cost of computational efficiency.

Single-shot detectors (SSDs) seek a middle ground. Rather than using a subnetwork to propose regions, SSDs rely on a set of predetermined regions. A grid of anchor points is laid over the input image, and at each anchor point, boxes of multiple shapes and sizes serve as regions. For each box at each anchor point, the model outputs a prediction of whether or not an object exists within the region and modifications to the box’s location and size to make it fit the object more closely. Because there are multiple boxes at each anchor point and anchor points may be close together, SSDs produce many potential detections that overlap. Post-processing must be applied to SSD outputs in order to prune away most of these predictions and pick the best one. The most popular post-processing technique is known as non-maximum suppression.

Finally, a note on accuracy. Object detectors output the location and label for each object, but how do we know how well the model is doing? For an object’s location, the most commonly-used metric is intersection-over-union (IOU). Given two bounding boxes, we compute the area of the intersection and divide by the area of the union. This value ranges from 0 (no interaction) to 1 (perfectly overlapping). For labels, a simple “percent correct” can be used.

Model architecture overview

R-CNN, Faster R-CNN, Mask R-CNN

A number of popular object detection models belong to the R-CNN family. Short for region convolutional neural network, these architectures are based on the region proposal structure discussed above. Over the years, they’ve become both more accurate and more computationally efficient. Mask R-CNN is the latest iteration, developed by researchers at Facebook, and it makes a good starting point for server-side object detection models.

YOLO, MobileNet + SSD, SqueezeDet

There are also a number of models that belong to the single shot detector family. The main difference between these variants are their encoders and the specific configuration of predetermined anchors. MobileNet + SSD models feature a MobileNet-based encoder, SqueezeDet borrows the SqueezeNet encoder, and the YOLO model features its own convolutional architecture. SSDs make great choices for models destined for mobile or embedded devices.

CenterNet

More recently, researchers have developed object detection models that do away with the need for region proposals entirely. CenterNet treats objects as single points, predicting the X, Y coordinates of an object’s center and its extent (height and width). This technique has proven both more efficient and accurate than SSD or R-CNN approaches.

Our model

the most suitable model for our project is MobileNet + SSD as SSD is faster than R-CNN because in R-CNN we need two shots one for generating region proposals and one for detecting objects whereas in SSD It can be done in a single shot.

The MobileNet SSD method was first trained on the COCO dataset and was then fine-tuned on PASCAL VOC reaching 72.7% mAP (mean average precision).

According to our hardware (jetson TX2) we are using jetson-inference , This repo uses NVIDIA TensorRT for efficiently deploying neural networks onto the embedded Jetson platform, improving performance and power efficiency using graph optimizations, kernel fusion, and FP16/INT8 precision.

Reference: 

1. Deep Learning authors: Ian Goodfellow, Yoshua Bengio, Aaron Courville