Operational parameters such as oil pressure are adjustable on each of TerramEarth's vehicles to increase their efficiency, depending on their environmental conditions. Your primary goal is to increase the operating efficiency of all 20 million cellular and unconnected vehicles in the field.
How can you accomplish this goal?
The best approach to increase the operating efficiency of all 20 million cellular and unconnected vehicles is to capture all operating data, train machine learning models that identify ideal operations, and run these models locally to make operational adjustments automatically. Given the significant number of vehicles that are unconnected, running models locally ensures that operational adjustments can be made without relying on a continuous internet connection. This approach allows each vehicle to adapt to its specific environmental conditions in real-time, thus optimizing efficiency across all vehicles.
B is correct. only 200k vehicle's are connected so need to run updates locally
In my view, option B says "run locally" referring to the machine learning models. "Machine learning models" is the subject of the sentence. Nowhere in the sentence says run "updates" locally. So running machine learning models would only make sense in Google's ML platform, not locally. Because of this reason, I believe the correct answer should be "D".
Both B and D starts with "Capture all operating data, train machine learning models that identify ideal operations, ..." so they are offering the same method for training the data. The keypoint here is "make operational adjustments" such as adjusting the oil pressure so if we host in GCP-ML, how are we going to instruct vehicles on field to adjust their oil pressure if they have no internet connection? There is no way to use GCP-ML model generated parameters to command the "not connected" field vehicles to make operational adjustments automatically. Therefore, I believe running it locally on the servers sitting in the vehicles is the only option. My answer: B
Making operational adjustments is an operational problem after recommendations are made by ML. In my mind, new data will keep feeding and total operational data changes every day for model and which would impact model performance over time. Monitoring model performance to achieve required efficiency levels would need some sort of centralization of efforts, as every machine environment condition might be different and there might be a need to create multiple models and test and operate them. (one shoe doesn't fit all).
B is ok
Making operational adjustments is an operational problem after recommendations are made by ML. In my mind, new data will keep feeding and total operational data changes every day for model and which would impact model performance over time. Monitoring model performance to achieve required efficiency levels would need some sort of centralization of efforts, as every machine environment condition might be different and there might be a need to create multiple models and test and operate them. (one shoe doesn't fit all).
Both B and D starts with "Capture all operating data, train machine learning models that identify ideal operations, ..." so they are offering the same method for training the data. The keypoint here is "make operational adjustments" such as adjusting the oil pressure so if we host in GCP-ML, how are we going to instruct vehicles on field to adjust their oil pressure if they have no internet connection? There is no way to use GCP-ML model generated parameters to command the "not connected" field vehicles to make operational adjustments automatically. Therefore, I believe running it locally on the servers sitting in the vehicles is the only option. My answer: B
Making operational adjustments is an operational problem after recommendations are made by ML. In my mind, new data will keep feeding and total operational data changes every day for model and which would impact model performance over time. Monitoring model performance to achieve required efficiency levels would need some sort of centralization of efforts, as every machine environment condition might be different and there might be a need to create multiple models and test and operate them. (one shoe doesn't fit all).
B is ok
What you think about D? there is a hint there :D and host in "Google Cloud Machine Learning" (ML)
how will the automatic opeartional adjustments be done to the unconnected/offline vehicles?
end of day
end of day
B is correct.
Making operational adjustments is an operational problem after recommendations are made by ML. In my mind, new data will keep feeding and total operational data changes every day for model and which would impact model performance over time. Monitoring model performance to achieve required efficiency levels would need some sort of centralization of efforts, as every machine environment condition might be different and there might be a need to create multiple models and test and operate them. (one shoe doesn't fit all).
B says "Capture all operating data". This is not right. You don't need ALL operating data to create an efficiency algorithm. The Answer is A.
how will the automatic opeartional adjustments be done to the unconnected/offline vehicles?
end of day
end of day
B)=> unconnected vehicles in the field.
Unconnected vehicles does not mean their data is not on GCP. you would still do ML on GCP and can use that to improve operational performance via maintenance port.
B. Train model in the cloud and deploy model to the edge for local prediction. This is typical in IoT.
The answer is B. Here is why: Capture all operating data (same in both) Train machine learning models that identify ideal operations (same in both) - We can train in Google ML as well but RUN it locally.The last sentence is different. 80% of vehicles are disconnected (as per case study) if we RUN (not train) the model in Google ML, how disconnected vehicles will (make operational adjustments automatically), in addition, the question mentioned (depending on their environmental conditions) which mean constantly changing conditions.
I will go with answer D A-> Engineers working on algorithm is not efficient way. B-> You can't run ML locally :) C-> Obviously NO. D-> Capture all data, run ML on this data, identify the ideal conditions. You can push adjustments either via IOT (for connected vehicles) or via maintenance ports for unconnected vehicles.
It doesn't require to run ML localy. It requires to apply operations locally (operation adjustements) and this is the main reason for voting B
You can run ML models locally. Sometimes you can even train ML models locally (federated ML)
B-> You can't run ML locally :) Says Who? You need GPU/TPU to train the model. Once model is trained you get MT file. You run your code on this MT file. You do not need to be online for this running MT file. You can run it in offline mode. So B is the correct answer.
I think it's D
I believe the catch is the word unconnected vehicles in the field. If this is the case, I would go with B
whether it is connected or unconnected, doesn't matter ... you gather all data and host it in Google Cloud Machine Learning (ML) Platform. fit for the question... I would go with D
I think the right answer id D), not B) The odd part of B) is "run locally". Who? Who is going to run locally? The dealer? Besides not talking anything about ML (maybe implicit), many architectures shows ML with a feed-back loop towards the IoT devices, *automatically*, closing the loop IMHO, it makes more sense.
B is correct. Training model is already running on GCP ML. The question is where to set up Inference. Setting up on GCP will not help unconnected vehicles. Setting up locally on all vehicles will meet our primary goal.
agree with you, it should be B, since they need to adjust the vehicles and it need to run in the vehicle itself, not in GCP.
This course about Google Cloud Platform, not about micro-controllers. It's not possible to run ML algorithms on tractors. And we already decided in previous questions, that we need to upgrade at least 80% of tractors for 4G connection. Answer D.
D is correct. B assumes that the vehicle computer can do ML jobs efficiently (i.e have GPU, or high-end CPUs, RAM etc).
those would be the requirements for "training", but not for "running" the model. I vote for B
B is the correct answer
B for me is possible, even if it require particular devices https://www.theverge.com/2020/1/14/21065141/google-coral-ai-edge-computing-products-applications-cloud But in this case the "connection to server" process is solved ...
D - you don't wish to waste any resources locally, all Machine Learning model training should occur in the cloud and vehicles should simply receive instructions to adjust operating parameters.
This question is about ML. So answer will be D. Every statement end with automatically so no body doing anything manual with the machines.
Answer is B; you still have to make the operational adjustments
It is possible to run a trained machine model locally. An example is your smartphone where it can perform offline real time language translation. So, I will go for B.
Answer is B
Answer : B Google has announced two new products aimed at helping customers develop and deploy intelligent connected devices at scale: Edge TPU, a new hardware chip, and Cloud IoT Edge, a software stack that extends Google Cloud’s powerful AI capability to gateways and connected devices. This lets you build and train ML models in the cloud, then run those models on the Cloud IoT Edge device through the power of the Edge TPU hardware accelerator. - https://cloud.google.com/blog/products/gcp/bringing-intelligence-edge-cloud-iot
from exam point of view, I doubt these questions would be part of new exam. Terram earth case study has changed.
B is correct. only 200k vehicle's are connected so need to run updates locally, means ML Edge
Yes, Exactly it is ML Edge.
I don't understand why everyone votes for B. In the replies I see, it seems to me that people are expecting the ML model to run on the vehicles and make changes offline. For this to be true, the vehicles would need a powerful computer, at least if this model is anywhere close to the AI models around... So, in this case, the ML model would have to be uploaded to all 20 million vehicles for them to change this parameters offline? That seems kind of crazy, but maybe I'm lacking some kind of info? Maybe the ML models are very portable and don't require much processing power?
Exactly. And considering the newly added data, the model would be updated, which would result in multiple versions of the model. Better alternative would be to centrally host the model and access it via API/offline via port
Training the model on ML platform, running the prediction model with live data as input on vehicle locally. I think it's B
"host in Google Cloud Machine Learning (ML) Platform" How unconnected vehicles communicate with GCML platform, only 200000 vehicle connected with cellular network. B is very close (no info on ability to run ml on edge)
The majority of the 20M vehicles are unconnected, so the correct answer is B.
B vs D I think D is grammarly better. An English test. Again.
By running locally I mean to install the trained model output (some equations with fine-tuned params) in devices itself to support off-grid devices as well. I think B is the correct answer here as per the requirement
Run Locally -> That's exactly what gcp ca, provide: https://cloud.google.com/blog/products/gcp/bringing-intelligence-edge-cloud-iot
B. Capture all operating data, train machine learning models that identify ideal operations, and run locally to make operational adjustments automatically
vote B
Also for me it's B. Maybe it might not seem so straightforward when reading the answer, but personally I understood it as 'running the result of ML training on devices'. Understanding it this way it certainly can only be B as from the scenario description we have most of the machines not connected to a cellular network and therefore it is impossible to pass data to them.
B. Capture all operating data, train machine learning models that identify ideal operations, and run locally to make operational adjustments automatically. Here's why: Capturing all operating data allows you to comprehensively understand vehicle performance under various conditions. Training machine learning models enables you to identify the ideal operational parameters for each vehicle based on environmental and operational conditions. Running these models locally on the vehicles allows real-time adjustments to be made, even for unconnected vehicles, enhancing operational efficiency without requiring constant communication with the cloud. This is especially important for vehicles that may not have a constant cellular connection.
The other options are less optimal because: A relies on manual rule creation, which is less flexible and less scalable than machine learning. C involves a streaming job and Google Cloud Messaging, which is not suitable for unconnected vehicles. D hosting models on the Google Cloud ML Platform requires constant connectivity, which isn't viable for unconnected vehicles.
I agree with B. Training in GCP ML Platform and run locally on unconnected devices to derive inferences is the right model.
I think that customers need the flexibility to run the ML model without connectivity. https://cloud.google.com/edge-tpu. B & D are good alternatives but i think that in this scenario is asking for a solution at the edge.
difference between B and D B) Without data analytics and machine learning, the two technologies just won't create meaningful algorithm for operational adjustments automatically. Besides, majority (99%) of the 20M vehicles are unconnected and the two technologies have to run on GCP for scalability so there is no way to communicate between local and GCP for adjustments automatically. (Correct Answer D) After creating good ML model by "Capture all operating data, train machine learning models that identify ideal operations", you can run the model in the vehicle to make operational adjustments automatically based on each specific vehicle's parameters. Probably run the model in the onboard computer or computer connected to the maintenance port.
small correction the answer is B once the models are created in GCP with bigquery as source(since the data it takes is from bigquery as per the design which is already populated with necessary data) you need to implement them locally since they are not connected to GCP.
There is very good explanation for B and D but processing has to be done in cloud ML and result is passed onto locally to implement it.so the answer is D.
Boils down between B and D. Only difference in wording is "run locally" for B and "host in Google Cloud Machine Learning (ML) Platform" for D. Both say, "make operational adjustments automatically". Multiple comments say B because the adjustments needs to be made local, this isn't about making adjustments, that's automatic in B and D according to the question. The question is about where to run or host the ML which in my opinion would NOT be done locally, but in the Google Cloud Machine Learning Platform. Considering that: D for me.
D is more relevant than B i guess..
D is more relevent than B i guess..
From previous Q's, a) not all vehicles are still connected b) but they are autonomous and each is equiped with server. so i would choose B. Why spending money to servers (not connecting vehicles, which is strange btw.) and after that spending more money to connect them and not using local vehicle resources.
Conditions in the questions changes and answer should be based upon the situation given. In exam you may not have the previous question and google is not even expecting you to know all the potential questions. Answer is D. ML is the right tool, while the adjustments are required based upon the conditions.
B is the Correct answer :)
D is ok
Sorry, I change to B. The question says that also disconnected vehicles should be optimized also
IMO - B is ok (because we have most vehicles unconnected)
Answer is B
B is right , Agreed with detailed discussions !
B is correct. only 200k vehicle's are connected so need to run updates locally
ok for B
Answer is B
It is D
Option B seems to be the most comprehensive approach. By capturing all operating data and training machine learning models to identify ideal operations, you can adaptively adjust operational parameters for each vehicle, thereby increasing overall efficiency. Hosting these models locally ensures real-time adjustments can be made autonomously without relying on external services or cloud infrastructure. The issue with option D is that it relies on hosting machine learning models in the Google Cloud Machine Learning Platform. While this could still potentially improve efficiency, it introduces additional latency and dependency on external cloud services, which might not be ideal for real-time operational adjustments required in the field. Additionally, hosting the models in the cloud could incur ongoing costs and potential connectivity issues, which might not be suitable for all operational environments.
I choose answer D. For answer B, running ML models locally on each vehicle is resource-intensive and might not be feasible for all vehicle types. It also makes updating and managing the models more difficult.
i think the issue here is that whether or not you want the model learn itself by continue provide the environment condition data . if you build and train the model in old data and deploy it local , will expect no continue learning by fitting new data.