The supply chain as we know it continues to evolve, and that is due in part to learning from the effects of the industry impacts over the last few years. As a result, more advanced supply chain technologies are focusing on the use of artificial intelligence primarily machine learning (ML) in various areas of their operations, and it is proving to be one of the most profound technologies enabling significant improvements in supply chains.

ML allows technology to teach itself over time, so that it can improve predictions, recommendations and decisions, but how does it relate to the supply chain exactly? In theory, it is nothing new, but in terms of the supply chain, it continues to prove beneficial in the advancement of digitization through data cleaning and supply chain planning, procurement and execution.

ML can deliver several benefits for supply chain management, one regarding an issue plaguing many companies today too much or too little inventory. The pandemic limited inventory due to backlogs, and now, many are finding they have too much inventory. With ML, organizations do not need to hold as much inventory because it optimizes the flow of products from one place to another. As a result, costs are reduced due to quality improvement and waste reduction, and products arrive in the marketplace just in time for sale as a result of upstream optimization.

Dealing with suppliers is one of the most challenging parts of supply chain management (SCM). With ML, supplier relationship management becomes easier due to simpler, proven administrative practices. ML can be implemented to analyze the types of contracts, documentation and other areas that lead to the best outcomes from suppliers and use those as a basis for future agreements and administration. Stakeholders get more insight into meaningful information, allowing for continual improvement and easier problem solving.

Quality is vital to good SCM as waste and faulty products create unnecessary rework and increase costs. ML can monitor how quality varies over time and suggest improvements. This doesnt just apply to materials and products. It can track other areas such as shipping, supplier and third-party quality.

ML isnt perfect, of course. It depends on reliable, high-quality and timely information, and lack of access to good data can cause significant issues. Supply chain managers need to have a robust approach to collecting and analyzing their data.

All organizations in the supply chain should provide information in a consistent way, and, where possible, SCM software should integrate with supplier and manufacturer systems to automatically collect and process data. There will need to be some human interaction with ML, especially with the quality of the data being collected. Supply chain information should be checked and audited periodically to ensure quality.

Machine learning models should be tested and checked to make sure outputs and suggestions are aligned with business needs and expectations.

For retailers, stock level analysis through ML can identify when products are declining in popularity and are reaching the end of their life in the retail marketplace. Price analysis can be compared to costs in the supply chain and retail profit margins to establish the best combination of pricing and customer demand. Also, upstream delays can be identified, allowing for contingency planning or alternative sourcing, and retailers can lower storage costs due to not having to hold as much stock.

Food manufacturers can use ML to conduct an analysis of commodity prices and weather patterns to optimize harvesting. Manufacturers can also increase speed to market by optimizing contracts and reducing turnaround times with upstream organizations.

The industry continues to focus on supply chain technologys role moving forward, and leaders will only benefit by riding the wave that ML is creating.

Glenn Jones is SVP of Product Marketing at Blume Global. He has a proven track record of growing businesses by building and leading R&D and product marketing organizations to define, develop, position and sell highly innovative and high value enterprise solutions delivered in the cloud. He was formerly the COO of Sweetbridge, the CTO of Steelwedge Software and also held leadership positions at other supply chain software companies including Elementum, E2Open and i2 Technologies.

Read this article:
How Machine Learning Optimizes the Supply Chain - Talking Logistics

A team of researchers at New York University has developed a model using machine learning that draws from news article content to predict locations that face food insecurity, according toNYU.

The model can help prioritize emergency food assistance allocation by helping uncover locations in most need.

Our approach could drastically improve the prediction of food crisis outbreaks up to 12 months ahead of time using both real-time news streams and a predictive model that is simple to interpret, said Samuel Fraiberger, a visiting researcher at NYU's Courant Institute of Mathematical Sciences, a data scientist at World Bank, and an author of the study, which appears in the journal Science Advances, in a statement.

Researchers collected text from more than 11 million news articles focused on roughly 40 food-insecure countries. They used their model to extract phrases and content related to food insecurity. Then they assessed the regions for food-insecurity factors like fatality counts, rainfall, vegetation, and food price changes, to determine a correlation between the news and food insecuritys impact. The high correlation indicated that the news stories served as an accurate indicator of the factors.

Traditional measurements of food insecurity risk factors, such as conflict severity indices or changes in food prices, are often incomplete, delayed, or outdated, said Lakshminarayanan Subramanian, a professor at the Courant Institute and one of the papers authors. Our approach takes advantage of the fact that risk factors triggering a food crisis are mentioned in the news prior to being observable with traditional measurements.

Related: Ukraine Export Helps Stabilize Cooking Oil Market; Giant Announces Sustainability Grant Program

Follow this link:
Machine Learning Helps Predict Food Crisis | News - Specialty Food Association

Artificial intelligence has become an integral part of our daily lives in today's technology-driven world. Although some people use neural networks and deep learning interchangeably, their advancements, features, and applications vary.

So what are neural networks and deep learning models, and how do they differ?

Neural networks, also known as neural nets, are modeled after the human brain. They analyze complex data, complete mathematical operations, look for patterns, and use the information gathered to make predictions and classifications. And just like the brain, AI neural networks have a basic functional unit known as the neuron. These neurons, also called nodes, transfer information within the network.

A basic neural network has interconnected nodes in the input, hidden, and output layers. The input layer processes and analyzes information before sending it to the next layer.

The hidden layer receives data from the input layer or other hidden layers. Then, the hidden layer further processes and analyzes the data by applying a set of mathematical operations to transform and extract relevant features from the input data.

It is the output layer that delivers the final information using the extracted features. This layer may have one or more nodes, depending on the data collection type. For binary classificationa yes/no problemthe output will have one node presenting a 1 or 0 result.

There are different types of AI neural networks.

Feedforward neural networks, mostly used for facial recognition, transfer information in one direction. This means every node in one layer is linked to every node in the next layer, with information flowing unidirectionally until it reaches the output node. This is one of the simplest types of neural networks.

This form of neural network aids theoretical learning. Recurrent neural networks are used for sequential data, like natural language and audio. They are also used for text-to-speech applications for Android and iPhones. And unlike feedforward neural networks that process information in one direction, recurrent neural networks use data from the procession neuron and send it back into the network.

This return option is critical for times when the system releases wrong predictions. Recurrent neural networks can attempt to find the reason for incorrect outcomes and adjust accordingly.

Traditional neural networks have been designed to process fixed-size inputs, but convolutional neural networks (CNNs) can process data of varying dimensions. CNNs are ideal for classifying visual data like images and videos of different resolutions and aspect ratios. They are also very useful for image recognition applications.

This neural network is also known as a transposed convolutional neural network. It is the opposite of a convolutional network.

In a convolutional neural network, input images are processed through convolutional layers to extract important features. This output is then processed through a series of connected layers, which carry out classificationassigning a name or label to an input image based on its features. This is useful for object identification and image segmentation.

However, in a deconvolutional neural network, the feature map that was formerly an output becomes the input. This feature map is a three-dimensional array of values and is unspooled to form the original image with an increased spatial resolution.

This neural network combines interconnected modules, each performing a specific subtask. Each module in a modular network consists of a neural network primed to tackle a subtask like speech recognition or language translation.

Modular neural networks are adaptable and useful for handling input with widely varying data.

Deep learning, a subcategory of machine learning, involves training neural networks to automatically learn and evolve independently without being programmed to do so.

Is deep learning artificial intelligence? Yes. It is the driving force behind many AI applications and automation services, helping users carry out tasks with little human intervention. ChatGPT is one of those AI applications with several practical uses.

There are many hidden layers between the input and output layers of deep learning. This allows the network to perform extremely complex operations and continually learn as the data representations pass through the layers.

Deep learning has been applied to image recognition, speech recognition, video synthesis, and drug discoveries. In addition, it has been applied to complex creations, like self-driving cars, which use deep learning algorithms to identify obstacles and perfectly navigate around them.

You must feed large amounts of labeled data into the network to train a deep-learning model. This is when backpropagation occurs: adjusting the weights and biases of the networks neurons until it can accurately predict the output for new input data.

Neural networks and deep learning models are subsets of machine learning. However, they differ in various ways.

Neural networks are usually made up of an input, hidden, and output layer. Meanwhile, deep learning models comprise several layers of neural networks.

Though deep learning models incorporate neural networks, they remain a concept different from neural networks. Applications of neural networks include pattern recognition, face identification, machine translation, and sequence recognition.

Meanwhile, you can use deep learning networks for customer relationship management, speech and language processing, image restoration, drug discovery, and more.

Neural networks require human intervention, as engineers must manually determine the hierarchy of features. However, deep learning models can automatically determine the hierarchy of features using labeled datasets and unstructured raw data.

Neural networks take less time to train, but feature lower accuracy when compared to deep learning; deep learning is more complex. Also, neural networks are known to interpret tasks poorly despite fast completion.

Deep learning is a complex neural network that can classify and interpret raw data with little human intervention but requires more computational resources. Neural networks are a simpler subset of machine learning that can be trained using smaller datasets with fewer computational resources, but their ability to process complex data is limited.

Though used interchangeably, neural and deep learning networks are different. They have different methods of training and degrees of accuracy. Nonetheless, deep learning models are more advanced and produce results with higher accuracy, as they can learn independently with little human interference.

View original post here:
Neural Networks vs. Deep Learning: How Are They Different? - MUO - MakeUseOf

Microsoft and Columbia Researchers Propose LLM-AUGMENTER: An AI System that Augments a Black-Box LLM with a Set of Plug-and-Play Modules  MarkTechPost

See more here:
Microsoft and Columbia Researchers Propose LLM-AUGMENTER: An AI System that Augments a Black-Box LLM with a Set of Plug-and-Play Modules -...

A New Deep Reinforcement Learning (DRL) Framework can React to Attackers in a Simulated Environment and Block 95% of Cyberattacks Before They Escalate  MarkTechPost

More here:
A New Deep Reinforcement Learning (DRL) Framework can React to Attackers in a Simulated Environment and Block 95% of Cyberattacks Before They Escalate...