The main focus of my current research is in the following three areas
· Network and Database Security
· Forecasting
· Algorithms and Software for Centralized Supply Chain Management


Network and Database Security   Top

There are many aspects of securing an enterprise - deployment and configuration of firewalls, user authentication/authorization, intrusion detection/prevention, etc. The case study of Application Service Provider (ASP) - based Supply Chain Management being developed offers numerous challenges in providing a secure computing environment.

In the most extreme case of resource sharing, orders, shipments and inventory levels of all suppliers and customers in the supply chain are stored on the ASP. Algorithms for SC optimization require some of a node's data to be visible to other partners in the chain. For example, the inventory of a product at retailer X should be visible to X's supplier, Y and also to Y's supplier of that product, etc. On the other hand, X may not wish its retail competitors to have access to such data.

Should DB access control be enforced in the application or database tiers?
In the traditional application development model, applications and the database are within the same boundary of trust. In this case, however, multiple business partners in the SC are permitted to host their custom-designed software on the ASP. For this reason, we propose a database tier firewall which filters or transforms each SQL query generated by the applications. In addition to implementation issues, there are a number of other concerns such as the specification/representation/updation of access constraints, translation of high-level access constraints to a level understood by the firewall and DBMS, performance implications, etc.

The above is intended to protect one client's data from the prying eyes of another (unauthorized) user. It does not protect the data from a malicious insider. Options towards the latter goal include encrypting the database or partly replicating the data at two or more unrelated sites.




Different forecasting techniques for a variety of applications such as retail sales, electric power consumption, share prices, etc. are experimented. We perform Holt-Winter forecasting (exponential smoothing of level, trend and seasonal index) which include yearly or monthly adaptation of smoothing constants over windows of varying size. These results are compared with those using the ARIMA (Auto-regressive Integrated Moving Average) model and neural networks.

We analyze the effect of decomposing a time series into trend, seasonal and irregular (noise) components. We perform a forecast on each series separately and then combine them to obtain a forecast on the original series.

Tens of forecasting experts, each proficient in one of the afore-mentioned techniques are identified. For a given series, a subset of experts is carefully chosen. Experts are assigned scores based on their performance. The scores are updated for each new point forecast. The new scores are used to "weigh" the opinions of the different experts in forecasting the next point. The choice of experts is frequently reviewed, so the best choice of experts for a given series varies dynamically.

Forecasting monthly sales data may help with strategic and tactical planning in SCM. However, from an operational perspective, forecasts of sales over the next 2-10 days are especially important in SCM. We are also actively investigating short-term forecasting of electric load (next-day forecasts of electricity consumption over 48 consecutive 30-minute intervals.) Here, three dimensions of seasonality are in evidence - during the day, week and year. We also factor the effects of temperature and humidity - these are positively correlated with power consumption in the Bombay area.

To address the numerous application domains, we are creating a high-performance forecast engine that can "talk" to multiple packages (such as SAS and others that we are building) and then use combining to obtain superior results.


Algorithms and Software for Centralized Supply Chain Management   Top

It has long been known that nodes in a supply chain that place orders based exclusively on local information observe a much larger variance in supplier order compared with variance of end customer demand. Moreover, the increased variability in order quantity propagates upstream in an amplified form leading to considerable mismatch between demand and supply (Bullwhip Effect).

Up-to-date sales data from retailers and accurate inventory levels at all downstream nodes from a given node are key to dampening the Bullwhip effect. In addition to these, updated lead times are crucial inputs to a forecast engine that computes re-order points and re-order levels. Under-estimation of re-order quantity or a delay in doing so will increase the stock-out probability. On the other hand, over-estimation of re-order quantity will increase holding cost.

Actually, the re-order quantity need not be determined by the tradeoff between stock-out and holding costs alone. In fact, it need not be determined by the customer! The trend toward Vendor Managed Inventory (VMI) puts the onus of maintaining adequate stock level on the supplier. The supplier (perhaps in collaboration with a third party logistics (3PL) provider) plans what, when and how much to supply to each of its customers with a view to minimizing the aggregate transportation, holding and stockout costs.

Easy 24 X 7 access to the internet promises an unprecedented level of information sharing throughout the supply chain. An attractive option is to store this information on an Application Service Provider (ASP). ASPs also provide computational power and disk space at a lower cost and remove technology risks (including obsolescence) out of investment decisions. Users are insulated from the pain of licensing, installation, upgrades, troubleshooting and other perfunctory tasks such as performing backups, recovery and maintenance.Our goal is to develop highly intelligent software to be hosted on the ASP that can provide a host of services for partners in the supply chain such as:

  • Demand/Sale forecasts at any node in the chain by analyzing long and short-term trends, seasonality, cyclicity, etc.
  • Triggers and alerts which inform entities of sudden surges or slumps in sales of different items
  • Computation of re-order points and quantities based on updated POS (point of sale) and lead time information
  • Payment gateway functionality to effect funds transfer from buyer to seller account by having direct connection to the banking network
  • Facilities to mine sales information to identify, for example, co-relations between sales of related items, the effect of promotions/discounts on sales volume/profits, etc.
  • Information regarding transport vehicle availability to the supplier and to the 3PL provider information regarding capacity planning, optimal routes/scheduling, delivery quantities (in the case of partial shipments), etc.

The object-oriented approach is ideally suited to ASP-based SCM. A typical node in the supply chain will include role-based components such as purchase, sales, inventory, transportation and forecasting departments with both, intra as well as inter-node linkages. Since the design will incorporate complex and possibly changing user requirements, there is much scope to exploit a number of structural and creational design patterns. Some of our clients may already be using an in-house ERP system. Advanced techniques in enterprise application integration would be needed in that case involving the use of integration patterns.

A model-driven architecture (using UML, OCL, etc.) would facilitate obtaining performance estimates - both, platform-independent and platform-specific - even without completing the first iteration of an implementation. This will enable us to derive and iterate through a design that is sensitive to performance from inception, not as an afterthought. We will also be able to study issues of practical interest such as load balancing, scalability and performance/security tradeoffs related to this application.