Local Bio-composites through AI-based formulation

[et_pb_section fb_built=”1″ _builder_version=”4.17.0″ _module_preset=”default” custom_padding=”0px||0px|||” global_colors_info=”{}”][et_pb_row _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.17.0″ _module_preset=”default” header_3_line_height=”1.1em” global_colors_info=”{}”]

Fungarium focuses on generating a range of materials with the premise of being local, bio and supply chain independent. How?

[/et_pb_text][/et_pb_column][/et_pb_row][/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”4.17.0″ _module_preset=”default” custom_padding=”0px|||||” global_colors_info=”{}”][et_pb_row _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”]

A true local potential

A harmful paradigm in our generation is the unsustainable supply chain. We at Fungarium foresee a shift in the paradigm of quality, especially in the younger generations. The idea of generating materials with processes where the carbon dioxide (CO2) added to the atmosphere is zero is a reality, even better some processes can extract CO2 from the atmosphere. The engineering of such processes will mark a turning point in our development as a civilization are crucial for the coming industrial revolution.

One of the biggest problems for generating products is the acquisition of high-quality materials from far away sources. For reasons of collection, processesing & transportation products leave a long trace of CO2 known as the carbon footprint. This carbon cost is high even for materials which are available locally, but due to specific traits, they are outsourced. We at fungarium decided to take a different path and we strive to produce the best of the locallity and bring long-term economic stability.

[/et_pb_text][/et_pb_column][/et_pb_row][/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_row column_structure=”1_2,1_2″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”1_2″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”]

Sourcing from around

Our pledge is to provide materials which are sourced from a specific radius. Which radius? According to a safe-net-zero emissions logistical radius, dependent on how far electric cargo vehicles can go collect and bring back to us.  This ensures that we are not polluting while collecting our materials. A platform allows us to account and be aware of seasonal and ready-to-take waste leftover from private and commercial farmers. The farmers pass over their “trash” which we transform into several mycelium-based materials.

[/et_pb_text][/et_pb_column][et_pb_column type=”1_2″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][/et_pb_column][/et_pb_row][/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_row column_structure=”1_2,1_2″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”1_2″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][/et_pb_column][et_pb_column type=”1_2″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”]

AI production

In recent years, the semiconducting industry has pioneered AI algorithms to allow a fine tuning of their best materials.1 This is done by simulating and experimenting hundreds of thousands of times. Although this approach is hard for computers, it has demonstrated to be a huge advantage. We at Fungarium take the same approach, given our experience, and look to optimize the values of density, elasticity, softness, etc. of our material by using ONLY local materials and allowing the AI algorithms to approximate or clearly identify the best combination of materials to obtain the customer specified trait.

To ensure that our AI is making the best out of our materials, our collaborations with THN in specific with the center of chemistry, materials, and production development, where we produce experimental data for our AI-training. We explicitly focus and materials from the region and for the region.

[/et_pb_text][/et_pb_column][/et_pb_row][/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_row _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”]

Want to go technical? Sure!

Let’s start by the beginning. Finding the ideal conditions satisfying multiple predefined targets simultaneously is a challenging decision-making process, which impacts all research. Additional complexity adds for tasks involving experimentation or expensive computations, as the number of evaluated conditions must be kept low. We propose an algorithm as a general purpose achievement scalarizing function for multi-target optimization where evaluations are the limiting factor.

Bayesian optimization approaches have emerged as an efficient alternative for such hurdles. Typically, they consist of two major steps: (i) a representation to the desired value of conditions (ii) a new set of conditions is proposed for the next evaluation based on this first samples. As such, this approach speculates about the experimental outcome using all previously conducted experiments, and verifies its speculations by requesting the evaluation of a new set of conditions. This new approach is easier for computation.

Bayesian optimization has gained increased attention as an alternative global optimization strategy as it was shown to reduce redundancy in the proposed conditions and, thus, locates global optima in fewer objective evaluations. It is a gradient-free strategy for the global optimization of possibly noisy black-box functions, which we denote with f from hereon. It consists of two major steps: (i) construct a surrogate to f and (ii) propose new parameter points for querying f based on this probabilistic approximation.

In the first step, the surrogate model is constructed by conditioning f on a prior φprior(θ) over the functional form, which is described by parameters θ. The parameters θ of the prior distributions are refined based on observations of n pairs Dn = (xk,fk)nk = 1 of parameter values xk, denoting, for instance, experimental conditions, and corresponding objective function values fk = f(xk), denoting the conditions set by our customers. The functional prior φprior is updated based on observations Dn to yield a more informative posterior φpost. With more and more observations Dn, the posterior φpost yields a better approximation and eventually converges to the objective function in the limit of infinitely many distinct observations, thus perfectly reproducing the experimental response landscape. This proxy model is used to propose new conditions for future evaluations via an acquisition function.

Popular in pharmaceutical industries, we aim to use Bayesian neural network (BNN) to estimate the parameter kernel density from the observed parameter points in an autoencoder-like architecture. As such, the BNN is used to nonlinearly estimate the density of the observed parameter points x we can construct an estimate to the parameter kernel density, which corresponds to a particular observed objective function value.

[/et_pb_text][/et_pb_column][/et_pb_row][et_pb_row _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_image src=”data:image/svg+xml;base64,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” _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][/et_pb_image][/et_pb_column][/et_pb_row][/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_row _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.17.0″ _module_preset=”default” global_colors_info=”{}”]

C. H. Chen, P. Parashar, C. Akbar, S. M. Fu, M. -Y. Syu and A. Lin, “Physics-Prior Bayesian Neural Networks in Semiconductor Processing,” in IEEE Access, vol. 7, pp. 130168-130179, 2019, doi: 10.1109/ACCESS.2019.2940130.

Robert Pollice, Gabriel dos Passos Gomes, Matteo Aldeghi, Riley J. Hickman, Mario Krenn, Cyrille Lavigne, Michael Lindner-D’Addario, AkshatKumar Nigam, Cher Tian Ser, Zhenpeng Yao, and Alán Aspuru-Guzik,”Data-Driven Strategies for Accelerated Materials Design”Accounts of Chemical Research 2021 54 (4), 849-860,doi: 10.1021/acs.accounts.0c00785

[/et_pb_text][/et_pb_column][/et_pb_row][/et_pb_section]

Leave a Comment

Your email address will not be published. Required fields are marked *