TensorFlow 2.0 shipped today, 30 September 2019, with new features, such as faster debugging and iteration with Eager Execution, a TensorFlow-enhanced implementation of the Keras API, and simplification and compatibility improvements across its APIs. TensorFlow 2.0 is a major upgrade that should increase performance, streamline workflows, and provide more compatibility for new or updated models.
We offer day 1 support
At Algorithmia, we believe data scientists should be able to deploy and serve models from any framework and keep up with the pace of tool development. To that end, we’re eager to announce that we support model deployments in the TensorFlow 2.0 framework—Google’s latest version that was released today.
Our Enterprise customers will receive the same support in their next product update.
While TensorFlow 2.0 includes a conversion tool for existing 1.x models, those conversions will not be fully automatic. Rest assured that the AI Layer will remain fully backward-compatible with all previous versions of TensorFlow—and the more than 15 other frameworks we support.
We won’t stop there. We want to provide users with the freedom to choose the best tool for every job, and that means immediate support for future versions of TensorFlow and other frameworks in development. If you have any questions about framework support or our rollout schedule, please contact your account manager or send us a message.
Happy model deployment!
While consumer-facing applications of machine learning (ML) have gotten a lot of attention (Netflix, Uber, and Amazon) the back office deserves some recognition. Enterprise-level systems that run the business—think finance, robo-advisors, accounting, operations, human resources, and procurement—tend to be large, complex, and process-centric. But, they also use and produce large amounts of both structured and unstructured data that can be handled in new ways to save time and money.
Machine learning combined with solution-specific software can dramatically improve the speed, accuracy, and effectiveness of back-office operations and help organizations reimagine how back-office work gets done.
A current trend among mid and large organizations is to implement Robotic Process Automation (RPA) in the back office to minimize manual tasks and achieve efficiencies. While there are specific use cases that make RPA an appropriate technology, there are significant differences with a machine learning approach.
Robotic Process Automation and artificial intelligence
Robotic Process Automation is software that mimics human actions while AI is built to simulate human intelligence. As an example, an RPA bot can be programmed to receive invoices via email (triggered on a specific subject line), download the invoice and put in a specific folder. An AI activity would be to “read” the invoice and extract the pertinent information like amount, invoice number, supplier name, due date, etc.
One of the more interesting downsides of RPA as outlined by Garter in the Magic Quadrant Report on RPA Vendors is that RPA automations create long-term technical debt, rather than overcoming it.
As you overlay RPA onto current technology and tasks, you are locking yourself into those technologies instead of updating and evolving.
Organizations must manually track the systems, screens, and fields that each automation touches in each third-party application and update the RPA scripts as those systems change. This is very challenging in a SaaS world in which product updates happen much more regularly than on-prem.
As such, the shift toward AI, and specifically ML, is to improve process, not just speed. Here are five specific applications of ML that can be used to improve back-office operations:
Account reconciliation (finance)
Account reconciliations are painful and error-prone. They are also critical to every business to ensure the proper controls are in place to close the books accurately and on-time. Many companies do this manually (which really means using Excel, macros, pivot tables, and Visual Basic) or have invested in RPA, which doesn’t get you very far, or in a Boolean rules-based system, which is expensive to set up and not super accurate.
Challenges in Account Reconciliation
An ML approach is ideal for account reconciliations, specifically matching reconciliations, because you have ground-truth data—previous successful matched transactions and consistent fields in subsequent reconciliations. The challenge has been that for large and complex data sets, the combinatorial problem of matching is really hard. Companies like Sigma IQ have focused on this problem and solved it with a combination of machine learning and point-solution software as a hosted platform.
Invoice processing/accounts payable (accounting)
We introduced invoice processing earlier in this article as a use case for ML in the back office as a way to understand the differences between RPA and ML. The reality is that every business deals with invoices at some level, and as natural language processing (NLP) and ML advance, these improvements will roll down from the enterprise level to small businesses.
Aberdeen Group indicates that well-implemented accounts payable systems can reduce time and costs by 75 percent, decrease error rates by 85 percent, and improve process efficiency by 300 percent, so it makes sense to pursue, right?
Using ML to augment accounting
Companies like AODocs are extending their NLP and ML capabilities to take some of the pain out of invoice management by automatically capturing information from invoices and triggering the appropriate workflow. These types of solutions can greatly reduce or eliminate manual data entry, increase accuracy, and match invoice to purchase order.
Employee attrition detection (HR)
There are many applications of AI in the HR function, including applicant tracking and recruiting (resume scanning and skills analysis), attracting talent before hiring, individual skills management/performance development (primarily via regular assessment analysis), and enterprise resource management.
Using ML to track employee satisfaction
One interesting use case from an ML−NLP perspective is employee attrition. Hiring is expensive, and retaining employees and keeping them happy is imperative to sustainable growth. Identifying attrition risk requires source data—like a consistently applied employee survey that uses unstructured data analysis for the open field comments. Overlaying this data with factors such as tenure, time since last pay raise/promotion, sick days used, scores on performance reviews, skill set competitiveness with market, and generally available employment market data can help assess probability of satisfaction.
Predicting repairs and upkeep for machinery (operations)
The influx of sensors into all types of equipment including trucks, oil rigs, assembly lines, and trains means an explosion of data on usage, wear, and tear of such equipment. Pairing this data with historical records on when certain types of equipment need certain preemptive maintenance means expensive machinery can be scheduled for downtime and repair not just based on number of hours used or number of miles driven, but what actual usage is.
Predix is a General Electric company that powers industrial apps to process the historic performance data of equipment. Its sensors and signals can be used to discern a variety of operational outcomes such as when machinery might fail so that you can plan for—or even prevent—major malfunctions and downtime.
Predictive analytics for stock in transit (procurement)
For companies that spend a lot of money on hard goods that need to be moved for either input into manufacturing or delivery to a retail shelf, stock in transit is a major source of opportunity for applying ML models to predict when goods will arrive at a destination.
Item tracking has improved dramatically with sensors, but it is only a point-in-time solution that doesn’t predict when the goods will arrive or when they should arrive. Weather, traffic, type of transport, risk probabilities, and historical performance are all part of the data that can help operations nail the flow of goods for optimal process timing.
Stock in Transit
SAP S/4HANA has an entire module dedicated to making trade-off predictions between different options for stock in transit solutions to meet customer order objectives.
Further opportunities for back office machine learning
These are just five of the hundreds of use cases ML paired with solution-specific software can be applied in order to improve the way the back office functions. Whether it is cutting down on manual tasks, improving accuracy, reducing costs, or helping teams change their critical processes wholesale, machine learning can augment nearly every back-office process.
The financial services industry has often been at the forefront of using new technology to solve business problems. It’s no surprise that many firms in this sector are embracing machine learning, especially now that increased compute power, network connectivity, and cloud infrastructure are cheaper and more accessible.
This post will detail five important machine learning use cases that are currently providing value within financial services organizations.
The cost of financial fraud for a financial services company jumped 9 percent between 2017 and 2018, resulting in a cost of $2.92 for every dollar of fraud. We have previously discussed machine learning applications in fraud detection in detail, but it’s worth mentioning some additional reasons why this is one of the most important applications for machine learning in this sector.
Most fraud prevention models are based on a set of human-created rules that result in a binary classification of “fraud” or “not fraud.” The problem with these models is that they can create a high number of false positives. It’s not good for business when customers receive an abnormally high number of unnecessary fraud notifications. Trust is lost, and actual fraud may continue to go on undetected.
Machine learning clustering and classification algorithms can help reduce the problem of false positives. They continually modify the profile of a customer whenever they take a new action. With these multiple points of data, the machine can take a nuanced approach to determine what is normal and abnormal behavior.
Creditworthiness is a natural and obvious use of machine learning. For decades, banks have used very rudimentary logistic regression models with inputs like income 30-60-90-day payment histories to determine likelihood of default, or the payment and interest terms of a loan.
The logistic model can be problematic as it can penalize individuals with shorter credit histories or those who work outside of traditional banking systems. Banks also miss out on additional sources of revenue from rejected borrowers who would likely be able to pay.
With the growing number of alternative data points about individuals related to their financial histories (e.g., rent and utility bill payments or social media actions), lenders are able to use more advanced models to make more personalized decisions about creditworthiness. For example, a 2018 study suggests that a neural network machine learning model may be more accurate at predicting likelihood of default as compared to logistic regression or decision-tree modeling.
Despite the optimism around increased equitability for customers and a larger client base for banks, there is still some trepidation around using black box algorithms for making lending decisions. Regulations, including the Fair Credit Reporting Act, require creditors to give individuals specific reasons for an outcome. This has been a challenge for engineers working with neural networks.
Credit bureau Equifax suggests that it has found a solution to this problem, releasing a “regulatory-compliant machine learning credit scoring system” in 2018.
Simply defined, algorithmic trading is automated trading using a defined set of rules. A basic example would be a trader setting up automatic buy and sell rules when a stock falls below or rises above a particular price point. More sophisticated algorithms exploit arbitrage opportunities or predict stock price fluctuations based on real-world events like mergers or regulatory approvals.
The previously mentioned models require thousands of lines of human-written code and have become increasingly unwieldy. Relying on machine learning makes trading more efficient and less prone to mistakes. It is particularly beneficial in high frequency trading, when large volumes of orders need to be made as quickly as possible.
Automated trading has been around since the 1970s, but only recently have companies had access to the technological capabilities able to handle advanced algorithms. Many banks are investing heavily in machine learning-based trading. JPMorgan Chase recently launched a foreign exchange trading tool that bundles various algorithms including time-weighted average price and volume-weighted average price along with general market conditions to make predictions on currency values.
Robo-advisors have made investing and financial decision-making more accessible to the average person. Their investment strategies are derived from an algorithm based on a customer’s age, income, planned retirement date, financial goals, and risk tolerance. They typically follow traditional investment strategies and asset allocation based on that information. Because robo-advisors automate processes, they also eliminate the conflict of financial advisors not always working in a client’s best interest.
While robo-advisors are still a small portion of assets under management by financial services firms ($426 billion in 2018), this value is expected to more than triple by 2023. Customers are enticed by lower account minimums (sometimes $0), and wealth management companies save on the costs of employing human financial advisors.
Cybersecurity and threat detection
Although not unique to the financial services industry, robust cybersecurity protocols are absolutely necessary to demonstrate asset safety to customers. This is also a good use case to demonstrate how machine learning can play a role in assisting humans rather than attempting to replace them. Specific examples of how machine learning is used in cybersecurity include:
Malware detection: Algorithms can detect malicious files by flagging never-before-seen software attempting to run as unsafe.
Insider attacks: Monitoring network traffic throughout an organization looking for anomalies like repeated attempts to access unauthorized applications or unusual keystroke behavior.
In both cases, the tedious task of constant monitoring is taken out of the hands of an employee and given to the computer. Analysts can then devote their time to conducting thorough investigations and determining the legitimacy of the threats.
It will be important to watch the financial sector closely because its use of machine learning and other nascent applications will play a large role in determining those technologies’ use and regulation across countless other industries.
Developing processes for integrating machine learning within an organization’s existing computational infrastructure remains a challenge for which robust industry standards do not yet exist. But companies are increasingly realizing that the development of an infrastructure that supports the seamless training, testing, and deployment of models at enterprise scale is as important to long-term viability as the models themselves.
Small companies, however, struggle to compete against large organizations that have the resources to pour into the large, modular teams and processes of internal tool development that are often necessary to produce robust machine learning pipelines.
Luckily, there are some universal best practices for achieving successful machine learning model rollout for a company of any size and means.
The Typical Software Development Workflow
Although DevOps is a relatively new subfield of software development, accepted procedures have already begun to arise. A typical software development workflow usually looks something like this:
This is relatively straightforward and works quite well as a standard benchmark for the software development process. However, the multidisciplinary nature of machine learning introduces a unique set of challenges that traditional software development procedures weren’t designed to address.
Machine Learning Infrastructure Development
If you were to visualize the process of creating a machine learning model from conception to production, it might have multiple tracks and look something like these:
It all starts with data.
Even more important to a machine learning workflow’s success than the model itself is the quality of the data it ingests. For this reason, organizations that understand the importance of high-quality data put an incredible amount of effort into architecting their data platforms. First and foremost, they invest in scalable storage solutions, be they on the cloud or in local databases. Popular options include Azure Blob, Amazon S3, DynamoDB, Cassandra, and Hadoop.
Often finding data that conforms well to a given machine learning problem can be difficult. Sometimes datasets exist, but are not commercially licensed. In this case, companies will need to establish their own data curation pipelines whether by soliciting data through customer outreach or through a third-party service.
Once data has been cleaned, visualized, and selected for training, it needs to be transformed into a numerical representation so that it can be used as input for a model. This process is called vectorization. The selection process for determining what’s important in the dataset for training is called featurization. While featurization is more of an art then a science, many machine learning tasks possess associated featurization methods that are commonly used in practice.
Since common featurizations exist and generating these features for a given dataset takes time, it behooves organizations to implement their own feature stores as part of their machine learning pipelines. Simply put, a feature store is just a common library of featurizations that can be applied to data of a given type.
Having this library accessible across teams allows practitioners to set up their models in standardized ways, thus aiding reproducibility and sharing between groups.
Current guides to machine learning tend to focus on standard algorithms and model types and how they can best be applied to solve a given business problem.
Selecting the type of model to use when confronted with a business problem can often be a laborious task. Practitioners tend to make a choice informed by the existing literature and their first-hand experience about which models they’d like to try first.
There are some general rules of thumb that help guide this process. For example, Convolutional Neural Networks tend to perform quite well on image recognition and text classification, LSTMs and GRUs are among the go-to choices for sequence prediction and language modeling, and encoder-decoder architectures excel on translation tasks.
After a model has been selected, the practitioner must then decide which tool to implement the chosen model. The interoperability of different frameworks has improved greatly in recent years due to the introduction of universal model file formats such as the Open Neural Network eXchange (ONNX), which allow for the porting of models trained in one library to be exported for use in another.
What’s more, the advent of machine learning compilers such as Intel’s nGraph, Facebook’s Glow, or the University of Washington’s TVM promise the holy grail of being able to specify your model in a universal language of sorts and have it be compiled to seamlessly target a vast array of different platforms and hardware architectures.
Model training constitutes one of the most time consuming and labor-intensive stages in any machine learning workflow. What’s more, the hardware and infrastructure used to train models depends greatly on the number of parameters in the model, the size of the dataset, the optimization method used, and other considerations.
In order to automate the quest for optimal hyperparameter settings, machine learning engineers often perform what’s called a grid search or hyperparameter search. This involves a sweep across parameter space that seeks to maximize some score function, often cross-validation accuracy.
Even more advanced methods exist that focus on using Bayesian optimization or reinforcement learning to tune hyperparameters. What’s more, the field has recently seen a surge in tools focusing on automated machine learning methods, which act as black boxes used to select a semi-optimal model and hyperparameter configuration.
After a model is trained, it should be evaluated based on performance metrics including cross-validation accuracy, precision, recall, F1 score, and AUC. This information is used to inform either further training of the same model or the next iterate in the model selection process. Like all other metrics, these should be logged in a database for future use.
Model visualization can be integrated at any point in the machine learning pipeline, but proves especially valuable at the training and testing stages. As discussed, appropriate metrics should be visualized after each stage in the training process to ensure that the training procedure is tending towards convergence.
Many machine learning libraries are packaged with tools that allow users to debug and investigate each step in the training process. For example, TensorFlow comes bundled with TensorBoard, a utility that allows users to apply metrics to their model, view these quantities as a function of time as the model trains, and even view each node in a neural network’s computational graph.
Once a model has been trained, but before deployment, it should be thoroughly tested. This is often done as part of a CI/CD pipeline. Each model should be subjected to both qualitative and quantitative unit tests. Many training datasets have corresponding test sets which consist of hand-labeled examples against which the model’s performance can be measured. If a test set does not yet exist for a given dataset, it can often be beneficial for a team to curate one.
The model should also be applied to out-of-domain examples coming from a distribution outside of that on which the model was trained. Often, a qualitative check as to the model’s performance, obtained by cross-referencing a model’s predictions with what one would intuitively expect, can serve as a guide as to whether the model is working as hoped.
For example, if you trained a model for text classification, you might give it the sentence “the cat walked jauntily down the street, flaunting its shiny coat” and ensure that it categorizes this as “animals” or “sass.”
After a model has been trained and tested, it needs to be deployed in production. Current practices often push for the deploying of models as microservices, or compartmentalized packages of code that can be queried through and interact via API calls.
Successful deployment often requires building utilities and software that data scientists can use to package their code and rapidly iterate on models in an organized and robust way such that the backend and data engineers can efficiently translate the results into properly architected models that are deployed at scale.
For traditional businesses, without sufficient in-house technological expertise, this can prove a herculean task. Even for large organizations with resources available, creating a scalable deployment solution is a dangerous, expensive commitment. Building an in-house solution like Uber’s Michelangelo just doesn’t make sense for any but a handful of companies with unique, cutting-edge ML needs that are fundamental to their business.
Fortunately, commercial tools exist to offload this burden, providing the benefits of an in-house platform without signing the organization up for a life sentence of proprietary software development and maintenance.
Algorithmia’s AI Layer allows users to deploy and serve models from any framework, language, or platform and connect to most all data sources. We scale model inference on multi-cloud infrastructures with high efficiency and enable users to continuously manage the machine learning life cycle with tools to iterate, audit, secure, and govern.
No matter where you are in the machine learning life cycle, understanding each stage at the start and what tools and practices will likely yield successful results will prime your ML program for sophistication. Challenges exist at each stage, and your team should also be primed to face them.
Natural language processing (NLP) is one of the fastest evolving branches in machine learning and among the most fundamental. It has applications in diplomacy, aviation, big data sentiment analysis, language translation, customer service, healthcare, policing and criminal justice, and countless other industries.
NLP is the reason we’ve been able to move from CTRL-F searches for single words or phrases to conversational interactions about the contents and meanings of long documents. We can now ask computers questions and have them answer.
Algorithmia hosts more than 8,000 individual models, many of which are NLP models and complete tasks such as sentence parsing, text extraction and classification, as well as translation and language identification.
Allen Institute for AI NLP Models on Algorithmia
The Allen Institute for Artificial Intelligence (Ai2), is a non-profit created by Microsoft co-founder Paul Allen. Since its founding in 2013, Ai2 has worked to advance the state of AI research, especially in natural language applications. We are pleased to announce that we have worked with the producers of AllenNLP—one of the leading NLP libraries—to make their state-of-the-art models available with a simple API call in the Algorithmia AI Layer.
Among the algorithms new to the platform are:
- Machine Comprehension: Input a body of text and a question based on it and get back the answer (strictly a substring of the original body of text).
- Textual Entailment: Determine whether one statement follows logically from another
- Semantic role labeling: Determine “who” did “what” to “whom” in a body of text
These and other algorithms are based on a collection of pre-trained models that are published on the AllenNLP website.
Algorithmia provides an easy-to-use interface for getting answers out of these models. The underlying AllenNLP models provide a more verbose output, which is aimed at researchers who need to understand the models and debug their performance—this additional information is returned if you simply set debug=True.
The Ins and Outs of the AllenNLP Models
Machine Comprehension: Create natural-language interfaces to extract information from text documents.
This algorithm provides the state-of-the-art ability to answer a question based on a piece of text. It takes in a passage of text and a question based on that passage, and returns a substring of the passage that is guessed to be the correct answer.
This model could feature into the backend of a chatbot or provide customer support based on a user’s manual. It could also be used to extract structured data from textual documents, such as a collection of doctors’ reports could be turned into a table that says (for every report) the patient’s concern, what the patient should do, and when they should schedule a follow-up appointment.
Entailment: This algorithm provides state-of-the-art natural language reasoning. It takes in a premise, expressed in natural language, and a hypothesis that may or may not follow up from. It determines whether the hypothesis follows from the premise, contradicts the premise, or is unrelated. The following is an example:
The input JSON blob should have the following fields:
- premise: a descriptive piece of text
- hypothesis: a statement that may or may not follow from the premise of the text
Any additional fields will pass through into the AllenNLP model.
The following output field will always be present:
- contradiction: Probability that the hypothesis contradicts the premise
- entailment: Probability that the hypothesis follows from the premise
- neutral: Probability that the hypothesis is independent from the premise
Semantic role labeling: This algorithm provides state-of-the-art natural language reasoning—decomposing a sentence into a structured representation of the relationships it describes.
The concept of this algorithm is considering a verb and the entities involved in it as its arguments (like logical predicates). The arguments describe who or what does the action of this verb, to whom or what it is done, etc.
NLP Moving Forward
NLP applications are rife in everyday life, and applications will only continue to expand and improve because the possibilities of a computer understanding written and spoken human language and executing on it are endless.