In this post I give an overview of the theory, tools, frameworks and best practices I have found until now around the testing (and debugging) of machine learning applications. I will start by giving an overview of the specificities of testing machine learning applications.
LINK
Both Software Engineering and Machine Learning have become recognized disciplines. In this article I analyse the combination of the two: engineering of machine learning applications. I believe the systematic way of working for machine learning applications is at certain points different from traditional (rule-based) software engineering. The question I set out to investigate is “How does software engineering change when we develop machine learning applications”?. This question is not an easy to answer and turns out to be a rather new, with few publications. This article collects what I have found until now.
LINK
While the technical application domain seems to be to most established field for AI applications, the field is at the very beginning to identify and implement responsible and fair AI applications. Technical, non-user facing services indirectly model user behavior as a consequence of which unexpected issues of privacy, fairness and lack of autonomy may emerge. There is a need for design methods that take the potential impact of AI systems into account.
DOCUMENT
In this proposal, a consortium of knowledge institutes (wo, hbo) and industry aims to carry out the chemical re/upcycling of polyamides and polyurethanes by means of an ammonolysis, a depolymerisation reaction using ammonia (NH3). The products obtained are then purified from impurities and by-products, and in the case of polyurethanes, the amines obtained are reused for resynthesis of the polymer. In the depolymerisation of polyamides, the purified amides are converted to the corresponding amines by (in situ) hydrogenation or a Hofmann rearrangement, thereby forming new sources of amine. Alternatively, the amides are hydrolysed toward the corresponding carboxylic acids and reused in the repolymerisation towards polyamides. The above cycles are particularly suitable for end-of-life plastic streams from sorting installations that are not suitable for mechanical/chemical recycling. Any loss of material is compensated for by synthesis of amines from (mixtures of) end-of-life plastics and biomass (organic waste streams) and from end-of-life polyesters (ammonolysis). The ammonia required for depolymerisation can be synthesised from green hydrogen (Haber-Bosch process).By closing carbon cycles (high carbon efficiency) and supplementing the amines needed for the chain from biomass and end-of-life plastics, a significant CO2 saving is achieved as well as reduction in material input and waste. The research will focus on a number of specific industrially relevant cases/chains and will result in economically, ecologically (including safety) and socially acceptable routes for recycling polyamides and polyurethanes. Commercialisation of the results obtained are foreseen by the companies involved (a.o. Teijin and Covestro). Furthermore, as our project will result in a wide variety of new and drop-in (di)amines from sustainable sources, it will increase the attractiveness to use these sustainable monomers for currently prepared and new polyamides and polyurethanes. Also other market applications (pharma, fine chemicals, coatings, electronics, etc.) are foreseen for the sustainable amines synthesized within our proposition.
Recycling of plastics plays an important role to reach a climate neutral industry. To come to a sustainable circular use of materials, it is important that recycled plastics can be used for comparable (or ugraded) applications as their original use. QuinLyte innovated a material that can reach this goal. SmartAgain® is a material that is obtained by recycling of high-barrier multilayer films and which maintains its properties after mechanical recycling. It opens the door for many applications, of which the production of a scoliosis brace is a typical example from the medical field. Scoliosis is a sideways curvature of the spine and wearing an orthopedic brace is the common non-invasive treatment to reduce the likelihood of spinal fusion surgery later. The traditional way to make such brace is inaccurate, messy, time- and money-consuming. Because of its nearly unlimited design freedom, 3D FDM-printing is regarded as the ultimate sustainable technique for producing such brace. From a materials point of view, SmartAgain® has the good fit with the mechanical property requirements of scoliosis braces. However, its fast crystallization rate often plays against the FDM-printing process, for example can cause poor layer-layer adhesion. Only when this problem is solved, a reliable brace which is strong, tough, and light weight could be printed via FDM-printing. Zuyd University of Applied Science has, in close collaboration with Maastricht University, built thorough knowledge on tuning crystallization kinetics with the temperature development during printing, resulting in printed products with improved layer-layer adhesion. Because of this knowledge and experience on developing materials for 3D printing, QuinLyte contacted Zuyd to develop a strategy for printing a wearable scoliosis brace of SmartAgain®. In the future a range of other tailor-made products can be envisioned. Thus, the project is in line with the GoChem-themes: raw materials from recycling, 3D printing and upcycling.
Carboxylated cellulose is an important product on the market, and one of the most well-known examples is carboxymethylcellulose (CMC). However, CMC is prepared by modification of cellulose with the extremely hazardous compound monochloracetic acid. In this project, we want to make a carboxylated cellulose that is a functional equivalent for CMC using a greener process with renewable raw materials derived from levulinic acid. Processes to achieve cellulose with a low and a high carboxylation degree will be designed.
