Medici Technologies provides predictive algorithm development and data analytics services, designed to enable companies to understand or transform their data into productive results. Medici accelerates client progress toward these goals using a structured approach that incorporates your understanding of the physical system while concurrently deploying our tools and methods in data analytics. Our tools include capabilities in regression analysis, classification analysis, signal processing, feature selection, clustering, design of experiments and statistical analysis.
Medici Technologies is a commercial entity singularly focused on creating value for our customers by leveraging our experience in the field of data analytics. Our goal is to provide a differentiated level of data processing capabilities to our customers through a combination of our scientific team, analysis tools and processing infrastructure. Our interdisciplinary team of scientists, engineers and programmers has historical experience and a strong education in data analytics. These individuals utilize the company’s Rapid Algorithm Assessment via Massive Parallelization system (aka RA2MP system). This system consists of a large existing code base of analytical tools and methods supported by a 500 node computer system. The system also contains associated databases, leaderboards and a messaging mechanism for storage and retrieval of results. This toolset allows Medici to provide clients with rapid results from algorithm assessment, optimization, robustness assessment via stress testing, and validation. This systematic approach solves difficult, complex, multivariate, life science algorithm problems in a deterministic manner.
Medici provides these data analytics services to smaller life science companies in a cost-effective manner. We look forward to working with you to address your data analysis needs.
Jump Start Program
Data analytics is a complex space with lots of acronyms, complicated processes, and a multitude of options. The Jump Start Program is designed to help both the client and Medici understand the problem and data, examine prior work, and define a plan for success. The goal of the Jump Start Program is to create real value through a defined scope and limited investment activity. The process allows a bidirectional exchange of information so the client can better understand the capabilities and effectiveness of the Medici approach. Conversely, the Jump Start process allows Medici to understand the applicability of our tools to your data. Please see the attachment on our Storefront for additional information.
The Jump Start Program is composed of the following 5 elements:
Exchange: On-Site Programmatic Review and Brainstorming Session
Medici realizes that in order to be effective in partnering with your organization, we need to quickly and effectively understand your project. To facilitate this transfer we:
• Conduct a multi-day visit at your facility so as to fully digest the problem, your data, and existing code, as well as brainstorm about what has worked, what you think will work, and our best guesses as to what might work
• Typically ask for completion of a standard list of interview style questions to guide information transfer so that the education process about the nature of your problem is thorough and efficient
• Create a SharePoint team site to securely house documents, data, and code repositories
Confirm: Validate Current Processing Method
The first step in improving your data analysis algorithms is to understand exactly what you’ve done and how well the process is working. The following steps allow Medici to fully understand the current data processing methodologies and the training process for predictive algorithms:
• Reproduce the code and results previously obtained on the project with clear communication of benchmark metrics
• Pragmatically estimate future performance using stringent segregation between training and validation data sets. Medici’s state-of-the-art training and assessment methods have proven to be the most reliable predictors of prospective performance available.
Experiment: Processing Method Perturbation Study
Once the performance of the existing methodologies has been defined, Medici can begin to explore ways to improve your processing methods. Using your ideas, in combination with the Medici RA2MP system, Medici can assess the level of performance improvement that might be realized by changing elements of the existing processing method. The Jump Start perturbation analysis will test 5 algorithm modifications that are viewed as likely to improve performance. The resulting changes in algorithm performance allow both parties insight into areas for improvement.
Challenge: Algorithm Stress Testing
Medici realizes that there is more to algorithm performance than a single performance metric calculated on a single data set. The RA2MP toolbox has a robustness, or stress testing, tool suite. The purpose of this testing is to identify performance variances or other vulnerabilities that may be present in the processing methodology. Our historical experience has shown that often the highest performing algorithm may not perform well across all anticipated real-world situations. For example, the Ferrari may be faster than the Volvo, but the Volvo is likely the better car for everyday use.
The client’s existing baseline processing method, as well as the 5 perturbations for the experiment section, will be stress tested. This 360 degree view will allow selection of the best possible algorithm for the client’s needs.
Report: Define Results and the Path Ahead
In addition to complete access to interactive OneNote notebooks and a designated SharePoint client site throughout the duration of the project, Medici will develop a final report detailing our findings and recommended next steps to move your product forward. Additionally, customers may request additional information, including associated literature, code reviews, data sufficiency reviews, test protocol assessment, experimental protocols and data collections plans.
The general objective of clustering is to group objects or data by similar characteristics. The process can be an important element in understanding the natural divisions within the data and can facilitate machine learning, pattern recognition, image analysis and other predictive methodologies.
The creation of meaningful results often requires careful examination or scrubbing of the data to remove errors, spurious results, outliers, or erroneous artifacts.
A critical step associated with any successful experimentation is to ensure that the reference determination is reliable, unbiased to the extent possible, and repeatable over the course of the study. Medici has significant experience in the examination of reference methodologies as well as experience in combining multiple methodologies for improved reference performance.
System Stability Assessment
In many experimental designs involving a multitude of parameters it is desirable to obtain test data at repeat/control conditions. These repeat conditions provide information regarding overall system stability and provide reliable quality control of the experimentation.
As many analytical tools have a fundamental assumption of linearity, data transformation to facilitate linear relationships is often conducted. Commonly used transformations include simple log transformation, Fourier transformation and wavelet transformation.
Medici’s RA2MP toolbox contains many Clustering and Classification algorithms that allow identification of distinct groups or classes in the data. Medici has experience with all aspects of classification, including two class and multi-class classifiers, linear and non-linear classifiers, and supervised, semi-supervised, and unsupervised classification problems. Medici also has tools to deal with class imbalance, in which there are very few examples of one of the classes, and tools to deal with incomplete or noisy class labels where the “true” label of some of the data might be unreliable.
Medici can accelerate classification model development via the use of the RA2MP toolbox. This toolbox allows Medici to leverage your understanding of the variables and systems under examination, test key assumptions, and quickly understand which aspects of the problem influence the performance of the classifier by leveraging massively parallel computing resources and designed experiments. Medici also has the tools and methods needed to stress test the final prediction method to look for potential weak points or unacceptable performance on sub-populations in your data.
Depending upon the noise structure of the data, additional signal processing, filtering or methods for noise reduction may be implemented. Data decomposition tools, also known as factoring, can be used for data compression and to project the data away from common noise sources. Commonly used factorization tools include PCA, ICA, and SELF.
The ability to effectively identify sources of signal versus noise artifacts is often a critical component of data analysis. Depending upon the purpose of the data, the methods can be remarkably successful at selecting a subset of relevant features to use for future data analysis or predictive model construction.
As it relates to the design of experiments, Medici has experience in fractional factorial designs and the subsequent use of these designs to create response surfaces, also known as surrogate models. Based upon customer needs, the resolution of the fractional design can be adjusted such that only main effects are identified or additional resolution can be utilized such that more subtle interactions are identified.
Monitoring Plan and Data Quality Control
Medici can assist in the development of the data monitoring plan and be an integral member of the team by providing real-time data assessment. Depending upon the scope of the project, such data monitoring activities can include statistical assessment, use of thresholds, and general outlier detection methodologies.
Many of the preceding steps are precursors to the development of effective predictive models. Predictive models are broadly defined as mathematical relationships used to examine measured information and to predict a desired result. Predictions may be quantitative and continuous in nature, discrete classification results, or combinations of both.
General Data Analysis Consulting
Medici Technologies has the interest and capabilities necessary to consult on your data analytics problem. Our experience base includes moderately simple due diligence activities, literature searches and reviews of results. More involved activities can include algorithm validation, brainstorming and idea development, intellectual property assessment, and benchmark metric development. More extensive engagements had been associated with significant data processing, support for product development, experimental oversight, and product improvement.
Medici has extensive experience using regression analysis to help customers understand the relationships amongst the variables in their data as well as using regression models to make predictions. The development of reliable, robust and high performing regression models requires a careful and systematic approach with attention given to all elements of the processing sequence. Key elements include: data representation, signal pre-treatment or processing, feature selection, regularization and model selection. Medici has experience with a wide variety of both linear and non-linear regression methods, including LASSO, GLM-NET, PLS, Support Vector Regression, MARS, and more.
Medici can accelerate regression model development via the use of the RA2MP toolbox. This toolbox allows Medici to leverage your understanding of the variables and systems under examination, test key assumptions, and quickly understand which aspects of the problem influence the performance of the regression model by leveraging massively parallel computing resources and designed experiments. Medici also has the tools and methods needed to stress test the final prediction method to look for potential weak points or unacceptable performance on sub-populations in your data.
The cornerstone activity following completion of the study is the analysis of variance (ANOVA). The standard approach involves a multifactor analysis with associated power calculations. Additional data processing and statistical analysis can be conducted as necessary.
Medici Technologies has not received any reviews.
Medici Technologies has not received any endorsements.