Inventory Pooling using Deep Reinforcement Learning
- Kameshwaran Sampath
- Sandeep Nishad
- et al.
- SCC 2022
I am currently with the Supply Chain group at IBM Research - India, Bangalore. My primary research interests are in AI, ML, and mathematical programming. My recent projects:
MEBN: Multi-Enterprise Business Networks
Industry 4.0’s digitization & integration of horizontal and vertical value chains would result in a large network of massively connected business enterprises, called as multi-enterprise business networks (MEBN). While a supply chain included trading partners of an enterprise, MEBN includes multi-tier trading partners spanning across multiple supply chains from different industries. Today’s supply chain and ERP systems restricts the visibility of an enterprise to its immediate Tier-1 trading partners. Around 51% of the supply chain disruptions originate with Tier-2 and Tier-3 suppliers and currently there is no early warning alert system for such disruptions. Due to Tier-1 visibility barrier, proactive contingency planning is non-optimal and reactive strategies are delayed and inefficient. MEBN breaks the Tier-1 visibility barrier with the network spanning across multiple tiers. This paradigm shift from enterprise supply chain to network of enterprises, creates interesting and relevant applications:
Multi-tier risk propagation
Control towers and orchestration
Identifying/recommending new partners
Network-wide comparative KPI of partners and business processes
The B2B ecosystems enabled by MEBN are further empowered with cross analytics from external data sources and news feeds.Our MEBN consists of tens and thousands of companies, with half a million business connections, across twenty different industry segments. With our AI capabilities including graph data science and probabilistic programming, and use of privacy enhancing techniques including zero knowledge proofs and secure multi party computation, we look at wide range of problems pertaining to anomaly and change detection, clustering, partner inference and recommendation, control towers and horizontal collaboration, etc.
Supply Chain ⊕ Blockchain
Blockchain, the underlying technology of cryptocurrencies, is applicable to many domains beyond finance and banking. Blockchain systems are, in essence, logically centralized but organizationally decentralized, without a central administrator. This perspective is helpful in identifying its synergy with supply chains, extending its application spectrum beyond visibility and tracking. In particular, our supply chain ⊕ Blockchain synergy focused on collaborative decision making, process orchestration, and support of non-deterministic optimization transactions in permissioned Blockchains.
Collaborative shipping refers to multiple independent shippers consolidating their freight flows to share the same set of carriers for goods transportation, in order to reduce cost. We design a collaborative shipping marketplace for spot shipments, where the shipping demands are bundled dynamically without premeditated planning by the shippers. Collaborative shipping requires a neutral third party as an orchestrator, who identifies the bundling opportunities, matches with the carrier, and allocates the cost among the shippers. Our Blockchain based application enables decentralized and transparent participation of shippers and carriers in choosing the matching of bundled demand with supply, without an orchestrator. A multi-disciplinary approach with techniques from multi criteria optimization, group decision making, cooperative game theory, and electoral voting are used to develop the distributed collaborative shipping application using Blockchain.
Process Orchestration of Invoice Reconciliation
As supply chains become more complex, more and more effort is being spent on reconciling differences between invoices of different parties and trading partners. Our Blockchain enabled solution bring end to end real-time shared visibility into the entire lifecycle of an invoice from the perspective of different stakeholders and systems in a trusted manner. The solution orchestrates the entire workflow of an Invoice Lifecycle using smart contracts and handles complex conflict prone events like Rejections and PO Amendments. Our pilot project with a leading global 2 and 3 wheeler manufacturing company successfully helped the company and its vendors drive operational efficiencies and improve tax compliance posture.
Optimization Transactions on Blockchains
A transaction is executed asynchronously by multiple nodes in a Blockchain and it is imperative that all executions result in the same output for the transaction to be valid. Transactions involving optimization problems can provide differing solutions on different executions. Common supply chain applications with planning, scheduling, resource allocation, and routing optimization are mixed integer linear programs with the above characteristic of generating different solutions. This inherent lack of determinism during the execution of optimization transactions impedes the adoption of Blockchain for the above supply chain applications. Our scheme enables atomic execution of optimization transactions in permissioned Blockchains like Hyperledger Fabric.
Dengue Modeling and Analytics
NEA: Prevention made with data. Made with IBM
Dengue has become a major international public health concern, particularly in countries in Southeast Asia, the Americas, Africa and Western Pacific. Dengue Fever is now endemic in more than 100 countries and about half the of the world's population is at risk. As a vector-borne viral disease, the spread of dengue is attributed to the expansion of the geographic distribution of the four serotypes of dengue viruses and their vector Aedes mosquitoes. The four dominant strains of dengue viruses have progressively spread to virtually all tropical countries around the globe. No specific vaccine or pharmaceutical treatment is available, so disease control is mostly based on prevention through the eradication of vector populations. In this project, we study the transmission dynamics of dengue using advanced spatio-temporal analytics and mathematical models of epidemiology. The key objectives would be to provide early warning of potential outbreaks of dengue disease for the purpose of implementing timely control measures. The spatio-temporal analytics takes a data centric approach using statistical and data mining models for prediction. The mathematical model of dengue transmission is a multi-population model that captures the transmission dynamics between host (human) and vector (mosquito) taking into account the four strains of dengue virus and the cross infections. We use the Spatiotemporal Epidemiological Modeler (STEM) to model and simulate dengue transmission.
IBM Optimization: Mine to Ship
Mining companies have complex supply chains that start from the mining location and stretch thousands of kilometers to the end customer in a different country and continent. The logistics of moving the materials from mines to ship is composed of series of optimization problems like berth allocation, ship scheduling, stockyard scheduling, and rail scheduling, which are individually NP-hard. We have developed an application, called as IBM Optimization: Mine to Ship, for end-to-end integrated operations scheduling. The application is built on IBM ILOG ODM Enterprise with advanced features like rescheduling under deviations and disturbances, and maintenance scheduling. The modeling and computational complexity of integrated scheduling optimization is tamed using hybrid optimization technique that leverages mathematical programming and constraint programming. The application will benefit the mining companies with increased resource usage, higher throughput, reduced cost of operations, and higher revenue. More details can be found here.