Amazon SageMaker: A Key to Accelerating Enterprise Machine Learning Adoption

On November 29th, Amazon Web Services announced SageMaker, a managed machine language service that manages the authoring, model training, and hosting of algorithms and frameworks. These capabilities can be used by themselves, or as an end-to-end production pipeline.

SageMaker is currently available with a Free tier providing 250 hours of t2.medium notebook usage, 50 hours of m4.xlarge training usage, and 125 hours of m4.xlarge hosting usage for hosting for two months. After two months or for additional hours, the service is billed per instance, storage GB, and data transfer GB.

Amalgam Insights anticipates watching the adoption of SageMaker as it solves several basic problems in machine learning.

Continue reading “Amazon SageMaker: A Key to Accelerating Enterprise Machine Learning Adoption”

Amazon Aurora Serverless vs. Oracle Autonomous Database: A Microcosm for The Future of IT

On November 29th, Amazon Web Services announced a variety of interesting database announcements at Amazon re:invent. Amazon Neptune, DynamoDB enhancements, and Aurora Serverless. Amalgam found both Neptune and DynamoDB announcements to be valuable but believes Aurora Serverless was the most interesting of these events both in its direct competition with Oracle and its personification of a key transitional challenge that all enterprise IT organizations face.

Amazon Neptune is a managed graph database service that Amalgam believes will be important for analyzing relationships, networked environments, process and anomaly charting, pattern sequencing, and random walks (such as solving the classic “traveling salesman” problem). Amazon Neptune is currently in limited preview with no scheduled date for production. Over time, Amalgam expects that Neptune will be an important enhancer for Amazon Kinesis’ streaming data, IoT Platform, Data Pipeline, and EMR (Elastic MapReduce) as graph databases are well-suited to find the context and value hiding in large volumes of related data.

For the DynamoDB NoSQL database service, Amazon announced two new capabilities. The first is global tables that will be automatically replicated across multiple AWS regions, which will be helpful for global support of production applications. Secondly, Amazon now provides on-demand backups for DynamoDB tables without impacting their availability or speed. With these announcements, DynamoDB comes closer to being a dependable and consistently governed global solution for unstructured and semistructured data.

But the real attention-getter was in the announcement of Aurora Serverless, an upcoming relational database offering that will allow end users to pay for database usage and access on a per-second basis. This change is made possible by Amazon’s existing Aurora architecture in separating storage from compute from a functional basis. This capability will be extremely valuable in supporting highly variable workloads.

How much will Aurora Serverless affect the world of relational databases?

Taking a step back, the majority of business data value is still created by relational data. Relational data is the basis of the vast majority of enterprise applications, the source for business intelligence and business analytics efforts, and the standard format that enterprise employees understand best for creating data. For the next decade, relational data will still be the most valuable form of data in the enterprise and the fight for relational data support will be vital in driving the future of machine learning, artificial intelligence, and digital user experience. To understand where the future of relational data is going, we have to first look at Oracle, who still owns 40+% of the relational database market and is laser-focused on business execution.

In early October, Oracle announced the “Autonomous Database Cloud,” based on Database 18c. The Autonomous Database Cloud was presented as a solution for managing the tuning, updating, performance driving, scaling, and recovery tasks that database administrators are typically tasked with and was scheduled to be launched in late 2017. This announcement came with two strong guarantees: 1) A telco-like 99.995% availability guarantee, including scheduled downtime and 2) a promise to provide the database at half the price of Amazon Redshift based on the processing power of the Oracle database.

In doing so, Oracle is using a combination of capabilities based on existing Oracle tuning, backup, and encryption automation and adding monitoring, failure detection, and automated correction capabilities. All of these functions will be overseen by machine learning designed to maintain and improve performance over time. The end result should be that Oracle Autonomous Database Cloud customers would see an elimination of day-to-day administrative tasks and reduced downtime as the machine learning continues to improve the database environment over time.

IT Divergence In Motion: Oracle vs. Amazon

In providing two very different next-gen database services, Oracle and Amazon have taken divergent paths in providing their next-generation relational databases. These decisions lead to an interesting head-to-head decision for companies seeking enterprise-grade database solutions.

On the one hand, IT organizations that are philosophically seeking to manage IT as a true service have, in Oracle, an automated database option that will remove the need for direct database and maintenance administration. Oracle is removing a variety of traditional corporate controls and replacing them with guaranteed uptime, performance, maintenance, and error reduction. This is an outcome-based approach that is still relatively novel in the IT world.

For those of us who have spent the majority of our careers handling IT at a granular level, it can feel somewhat disconcerting to see many of the manual tuning, upgrading, and security responsibilities being both automated and improved through machine learning. In reality, highly repetitive IT tasks will continue to be automated over time as the transactional IT administration tasks of the 80s and 90s finally come to an end. The Oracle approach is a look towards the future where the goal of database planning is to immediately enact analytic-ready data architecture rather than to coordinate efforts between database structures, infrastructure provisioning, business continuity, security, and networking. Oracle has also answered the question of how it will answer questions regarding the “scale-out” management of its database by providing this automated management layer with price guarantees.

In this path of database management evolution, database administrators must be architects who focus on how the wide variety of data categories (structured, semi-structured, unstructured, streaming, archived, binary, etc…) will fit into the human need for structure, context, and worldview verification.

On the other hand, Amazon’s approach is fundamentally about customer control at extremely granular levels. Aurora is easy to spin up and allows administrators a great deal of choice between instance size and workload capacity. With the current preview of Amazon Aurora Serverless, admins will have even more control over both storage and processing consumption by starting at the endpoint level as a starting point for provisioning and production. Amazon will target the support of MySQL compatibility in the first half of 2018, then follow with PostgreSQL later in 2018. This billing will occur in Aurora Capacity Units as a combination of storage and memory metered in one-second increments. This granularity of consumption and flexibility of computing will be very helpful in supporting on-demand applications with highly variable or unpredictable usage patterns.

But my 20+ years in technology cost administration also lead me to believe that there is an illusory quality of control in the cost and management structure that Amazon is providing. There is nothing wrong with providing pricing at an extremely detailed level, but Amalgam already finds that the vast majority of enterprise cloud spend unmonitored from a month-to-month basis at all but the most cursory levels. (For those of you in IT, who is the accountant or expense manager who cross-checks and optimizes your cloud resources on a monthly basis? Oh, you don’t have one?)

Because of that, we at Amalgam believe that additional granularity is more likely to result in billing disputes or complaints. We will also be interested in understanding the details of compute: there can be significant differences in resource pricing based on reserved instances, geography, timing, security needs, and performance needs. Amazon will need to reconcile these compute costs to prevent this service from being an uncontrolled runaway cost. This is the reality of usage-based technology consumption: decades of telecom, network, mobility, and software asset consumption have all demonstrated the risks of pure usage-based pricing.

Amalgam believes that there is room for both as Ease-of-Use vs. Granular Management continues to be a key IT struggle in 2018. Oracle represents the DB option for enterprises seeking governance, automation, and strategic scale while Amazon provides the DB option for enterprises seeking to scale while tightly managing and tracking consumption. The more important issue here is that the Oracle DB vs. Amazon DB announcements represent a microcosm of the future of IT. In one corner is the need to support technology that “just works” with no downtime, no day-to-day adminstration, and cost reduction driven by performance. In the other corner is the ultimate commoditization of technology where customers have extremely granular consumption options, can get started at minimal cost, and can scale out with little-to-no management.

Recommendations

1) Choose your IT model: “Just Works” vs. ” Granular Control.” Oracle and Amazon announcements show how both models have valid aspects. But inherent in both are the need to both scale up and scale out to fit business needs.

2) For “Just Works” organizations, actively evaluate machine learning and automation-driven solutions that reduce or eliminate day-to-day administration. For these organizations, IT no longer represents the management of technology, but the ability to supply solutions that increase in value over time. 2018 is going to be a big year in terms of adding new levels of automation in your organizations.

3) For “Granular Control” organizations, define the technology components that are key drivers or pre-requisites to business success and analyze them extremely closely. In these organizations, IT must be both analytics-savvy and maintain constant vigilance in an ever-changing world. If IT is part of your company’s secret sauce and a fundamental key to differentiated execution, you now have more tools to focus on exactly how, when, and where inflection points take place for company growth, change, or decline.

Why Did TIBCO Acquire Alpine Data?

On November 15, 2017, TIBCO announced the acquisition of Alpine Data, a data science platform long known for its goals of democratizing data science and simplifying access to data, analytic workflows, parallel compute, and tools.

With this acquisition, TIBCO makes its second major foray into the machine learning space after June 5th acquisition of Statistica. In doing so, TIBCO has significantly upgraded its machine learning support capabilities, which will be especially useful to TIBCO in continuing to position itself as a full-range data and analytics solution.

When this acquisition occurred, Amalgam received questions on how Alpine Data and Statistica would be expected to work together and how Alpine Data would fit into TIBCO’s existing machine learning and analytics portfolio. Amalgam has provided favorable recommendations for both Alpine Data and Statistica in 2017 and plans to continue providing a positive recommendation for both solutions, but sought to explore the nuances of these recommendations.

In our Market Milestone, we explore why Alpine Data was a lower-ranked machine learning solution in analyst landscapes despite being early-to-market in providing strong collaborative capabilities and supporting a wide variety of data sources. We also wanted to explore the extent to which Alpine Data provided some sort of conflict to existing TIBCO customers. Finally, we also wanted to provide guidance on how TIBCO’s acquisition would potentially change Alpine Data’s positioning and capabilities.

To read Amalgam Insights’ view and recommendations regarding this report, use the following link to acquire this report.

How Does myTrailhead Excel and How Can Science Make myTrailhead Even Better

Astro, Einstein, and other Salesforce Trailhead characters
Salesforce’s Trailhead branding is both on-point and adorable.

Key Takeaways:

  • myTrailhead allows customized training content and incorporates useful motivational and performance testing tools.
  • myTrialhead could be enhanced by incorporating scientifically-validated best practices in training, which suggest that hard skills are best trained by a cognitive skill learning system in the brain and soft skills are best trained by a behavioral skill learning system in the brain
  • In its current implementation, myTrailhead is more nearly optimized for hard skill training, but is sub-optimal for soft skills training

Technology is progressing at an accelerating rate. Jobs are constantly being updated or redefined by, and with the help of technology. Employees are constantly being asked to learn new skills whether in the same job or in a new position. Constant training is the rule, not the exception, and training platforms must be built with this in mind.
Continue reading “How Does myTrailhead Excel and How Can Science Make myTrailhead Even Better”

Why Computer Based Sexual Harassment Awareness Training is Ineffective and Why Virtual Reality (VR) Offers a Better Solution

Person participating in virtual reality simulation
Photo by Samuel Zeller

What do Bill O’Reilly, Harvey Weinstein, Roy Moore, Louis C.K., Michael Oreskes, Kevin Spacey and many others have in common (other than all being male)? Certainly not political beliefs or professional expertise. Whether left or right leaning in the political arena, or focused on entertainment, journalism or government service, all have been accused of, and in some cases admitted to, sexual harassment or sexual assault.

Sexual misconduct has been a cancer on society for as long as history has been recorded. Have we reached a tipping point? Has the “#MeToo” movement and the press coverage led to a fundamental shift in our thinking and will it permanently affect behavior? These are great questions, that I am not qualified to answer. Only time will tell.

What I am prepared to say is that our approach to sexual harassment awareness, in particular, training programs focused specifically on increasing awareness of sexual harassment and reducing the incidence of sexual harassment, are nearly all sub-optimal.

Why?

Computer-Based Sexual Harassment Awareness Training is Sub-Optimal

Whether developed for government or corporate entities large and small, nearly all sexual harassment awareness training programs are classroom or computer-based. They involve having individuals read text, or watch slideshows and videos that define sexual harassment and the behaviors that are appropriate or inappropriate. They describe power differentials that often exist in government or the corporate world and how that impacts the appropriateness of interpersonal interactions. They might even include video interactions so that individuals can “see” sexual harassment in action from a third-person perspective.

In all of these cases, the nature of the training content and the training procedures are such that they recruit the cognitive skills learning system in the brain. The cognitive skills learning system in the brain learns through observation, mimicry and mental repetition. This is an excellent brain system for learning hard skills such as: (a) learning new software, (b) becoming proficient with rules and regulations, or (c) learning a new programming language, but this learning system in the brain is less effective for learning soft skills such as appropriate interpersonal interactions and real-time communication, or for training true empathy for another’s situation.

Appropriate interpersonal interactions and real-time communication skills are best learned by the behavioral skills learning system in the brain that learns by doing and receiving immediate corrective feedback. Physical repetitions, not mental repetitions, are key. Genuine empathy for another’s situation is best trained through a first-person experience in which you “are” that other person.

The Promise of VR for Sexual Harassment Awareness Training

VR offerings currently come in two general types. One takes a first-person perspective and allows you to literally “walk a mile” in someone else’s shoes. This approach involves passive, observational learning, much like computer based training, but the feeling of immersion, and more importantly the feeling that you are “someone else” is powerful. I believe that this offers one of the most effective tools for enhancing emotional intelligence and helping learners understand at a visceral level what it is like to be in a position of weakness and to be the direct target of sexual harassment. There is no better way for a middle-aged, Caucasian male to “feel” the prejudice or sexual harassment that a young, female African-American might experience or to “feel” the discrimination that many members of the LGBT community feel, than to put that man in a first-person VR environment where they are that other individual. Of course, the training content and the training scenarios must be realistic to be effective, but experts in this sector know how to create high-quality content. In my view, first-person VR experiences offer a great first step toward reducing the incidence of sexual harassment by increasing genuine empathy and understanding.

Although these passive, observational VR experiences offer a great tool for enhancing sexual harassment awareness, they are not focused specifically on behavior. The second type of VR offering, interactive VR, addresses this problem directly. Interactive VR platforms incorporate realistic interpersonal interaction and real-time communication into the mix. The learner can be placed in situations involving sexual harassment in which virtual agents react to the learner’s behavior in real-time. In other words, learners learn by doing and by receiving immediate feedback regarding the correctness of their behavior. This approach optimally recruits the behavioral skills learning system in the brain, which is the ideal system for reducing the incidence of inappropriate behaviors. Without taking a deep dive into brain neurochemistry, suffice it to say that behavioral skills learning is best when the brain circuits that initiated the behavior are still active when feedback is received. If the action is appropriate, then that behavior will be strengthened, and if the action is inappropriate, then that behavior will be weakened. Although there are clearly ethical limits to the intensity of the VR environments that one can be compelled to experience, interactive VR experiences with even mild levels of harassment will be effective in changing behavior.

Interactive VR approaches may also be useful in extreme cases as a rehabilitation procedure. Individuals already identified as sexual harassers by previous actions or complaints may benefit significantly from this type of rehabilitative behavioral therapy. In these situations, it may be ethically appropriate to increase the intensity of the interactive VR environments so that real changes in behavior will occur.

Disclaimer

Sexual harassment is a serious problem in our society. In many cases, the individual is fully aware of their behavior and simply does not care. In such cases, no training, whether computer-based or VR, will likely have any effect. These are situations involving a conscious bias and behavioral change may be difficult. It is the cases of unconscious bias, where the individual is less aware of the impact of their behavior, that there is hope. The point of this article is not to claim that all sexual harassment can be eradicated. That is unrealistic, wishful thinking. That said, I believe that we can reduce the incidence of sexual harassment through effective training. I believe that the science of learning suggests that VR may provide a better tool for achieving this goal than computer based training.

Conclusion

I am not an expert on sexual harassment, but I do understand the psychology of behavior and behavior change. Although traditional computer-based approaches do their best to define, describe and demonstrate sexual harassment behavior, they target the cognitive skills learning system in the brain. This system is ideal for hard skill training, but not soft skill training, such as the training needed to reduce the incidence sexual harassment. I believe that VR holds significant promise as a training tool for reducing the incidence of sexual harassment. By combining the passive, observational first-person VR experiences that allow one to see the world through someone else’s eyes and experience sexual harassment first hand, with interactive VR experiences that allow one to engage in interpersonal interaction and real-time communication focused on rewarding appropriate behaviors and punishing inappropriate behaviors we might be able to reduce the size of this cancer from our society. The science is clear, and it suggests that this VR approach has merit.

With myEinstein, Salesforce Embraces that “AI is the New UI”

Astro, Einstein, and other Salesforce Trailhead characters
Salesforce Einstein Airplane - Courtesy of Salesforce
Salesforce Einstein Airplane – Courtesy of Salesforce

Key Takeaway: Amalgam believes that the go-live date of myEinstein will be the most important date for Enterprise AI in 2018 as it represents the day that AI will become practical and available to a broad business audience across industries, verticals, company sizes, and geographies.

On November 6, 2017, Salesforce [NYSE:CRM] announced the launch of myEinstein: services based on Salesforce’s Einstein machine learning platform to support point-and-click-based and codeless AI app development. This announcement was one of several new services that Salesforce built across platform (mySalesforce and myIoT), training (myTrailhead), and user interface development (myLightning).

myEinstein consists of two services: Continue reading “With myEinstein, Salesforce Embraces that “AI is the New UI””

4 Key Developer Responsibilities Where Machine Learning Can Help

Note: A version of this post was published to Tom’s Tech Take II

As the fall season of tech conferences starts to wind down, something is quite clear – machine learning (ML) is going to be everywhere. Box is putting ML in content management, Microsoft in office and CRM, and Oracle is infusing it into, well, everything. While not a great leap forward on the order of the Internet, smartphones, or PCs, the inclusion of ML technology into so many applications will make a lot of mundane tasks easier. This trend promises to be a boon for developers. The strength of machining learning is finding and exploiting patterns and anomalies. What could be more useful to developers?

Here are some examples:

Please register or log into your Free Amalgam Insights Community account to read more.
Log In Register

Amalgam Insights Analyzes Sage Intacct and Pacioli AI

Amalgam Insights recently attended Sage Intacct Advantage. In the past, Intacct got AI’s attention for its strong technology foundation that positions it well for a future of predictive analytics, ease of integration, and machine learning while maintaining the core financial responsibilities associated with being a nominative mid-market ERP solution. Sage has traditionally been known as…

Please register or log into your Free Amalgam Insights Community account to read more.
Log In Register

With Oracle Universal Credits, the Cloud Wars Are Truly On

In late September, prior to Oracle Open World, Oracle (NYSE: ORCL) held an event to announce its consumption pricing model of Universal Credits and the ability to reuse existing software licenses across Oracle’s Platform as a Service (PaaS) middleware, analytics, and database offerings. The Universal Credits represent a fundamental change in cloud pricing as they will allow Oracle Cloud customers to switch between Oracle’s IaaS and PaaS services. In addition, Larry Ellison also unveiled a “self-driving” database that would greatly reduce the cost of administration.

Please register or log into your Free Amalgam Insights Community account to read more.
Log In Register

28 Hours as an Industry Analyst at Strata Data

grid-725269_640
grid-725269_640
grid-725269_640

Last week, I attended Strata Data Conference at the Javitz Center in New York City to catch up with a wide variety of data science and machine learning users, enablers, and thought leaders. In the process, I had the opportunity to listen to some fantastic keynotes and to chat with 30+ companies looking for solutions, 30+ vendors presenting at the show, and attend with a number of luminary industry analysts and thought leaders including Ovum’s Tony Baer, EMA’s John Myers, Aberdeen Group’s Mike Lock, and Hurwitz & Associates’ Judith Hurwitz.

From this whirwind tour of executives, I took a lot of takeaways from the keynotes and vendors that I can share and from end users that I unfortunately have to keep confidential. To give you an idea of what an industry analyst notes, following are a short summary of takeaways I took from the keynotes and from each vendor that I spoke to:

Keynotes: The key themes that really got my attention is the idea that AI requires ethics, brought up by Joanna Bryson, and that all data is biased, which danah boyd discussed. This idea that data and machine learning have their own weaknesses that require human intervention, training, and guidance is incredibly important. Over the past decade, technologists have put their trust in Big Data and the idea that data will provide answers, only to find that a naive and “unbiased” analysis of data has its own biases. Context and human perspective are inherent to translating data into value: this does not change just because our analytic and data training tools are increasingly nuanced and intelligent in nature.

Behind the hype of data science, Big Data, analytic modeling, robotic process automation, DevOps, DataOps, and artifical intelligence is this fundamental need to understand that data, algorithms, and technology all have inherent biases as the following tweet shows:

Please register or log into your Free Amalgam Insights Community account to read more.
Log In Register