What is the name of the ion that increases in concentration when an acid is added to water?A. Hydroxide IonB. Hydronium IonC. Oxide IonD. Buffer Ion

How do hydrogen bonds affect the amount of time it takes for water to heat up? A. Polar molecules have fewer hydrogen bonds which allows them to move quickly, causing them to slowly heat upB. Non-polar and polar molecules have the same amount of hydrogen bonds which causes them to slowly heat upC. Polar molecules have too many packed together which causes them to slowly heat upD. Polar molecules need hydrogen bonds to be broken up in order for them to heat up, which causes water to slowly heat upE. Non-polar molecules have hydrogen bonds that easily break up, which makes them heat up quickly compared to polar molecules

What is the primary factor determining whether an atom will bond with another atom to achieve a stable electron configuration?A. The number of neutrons in the atomB. The amount of heat energy available to the atomsC. The number of valence electrons in the atomD. The size of the atomic nucleus

Bonus question (5 points) ref: OpenAI’s Sora, using the prompt — “Create two videos in which an economics professor teaches the economics of AI, one with a smiling face and one without.” First, generate a base image of the professor. Second, create a video without a smile. Then, generate the smiling version of the video using additional prompts. The entire process takes approximately 5–10 minutes. An economics professor wants to estimate the effect of smiling during online lectures on students’ academic performance. To mitigate omitted variable bias and confounding factors, the professor employs a randomized controlled trial (RCT). Prior to implementing the RCT, the professor uses AI to generate counterfactual images and videos of themselves smiling. Students in the control group are randomly assigned to watch recorded lectures or view images of the professor without a smile. Students in the treatment group are randomly assigned to watch the same lectures or view images of the professor with a smile (generated using AI). The goal is to estimate the counterfactual treatment effect of the professor’s smile on students’ learning outcomes in an online setting based on the following linear regression equation:

______________ means the gradient of deep learning models during training is too small or too large (unstable) ; therefore, the deep learning model cannot learn more effectively. To mitigate this problem, we can use the ReLU function as an activation function, use gradient clipping to normalize it when the norm of the gradient is larger than the threshold, and add gate structures in the deep learning model to consider long-term dependency in the dataset.

Ref: https://huggingface.co/learn/nlp-course  Parallel computation is applicable for transformer model; therefore, it is easy to increase the size of the model with a larger corpus data. Language model (LM) and large language model (LLM) are mostly based on transformer model. LM is divided into three groups: encoder-only model, decoder-only model, and encoder-decoder model. For example, the Bidirectional Encoder Representations from Transformers (BERT) model can be used to generate embedding vectors from text data for text classification, because BERT is (1)__________________( a. encoder-only model, b. decoder-only model). For example, OpenAI’s ChatGPT is based on Generative Pre-trained Transformer (GPT) 4 and 4.5. GPT models are more suitable for text generation in diverse tasks, because GPT is a (2)__________________( a. encoder-only model, b. decoder-only model).

A convolutional neural network (CNN) is widely used for text and image classification. CNN models contain a convolutional and pooling layer. A convolutional layer have (1)___________ for a convolutional operation as follows: (2)__________ determines the number of pixels by which we move (1)_________ across the input matrix as follows: (3)_________ is adding a boundary around the input matrix to preserve the size of the matrix in the output as follows:   Ref: https://github.com/WegraLee/deep-learning-from-scratch?tab=readme-ov-file Please select the right answer _________________ for (1), (2), and (3)      

Ref: https://github.com/WegraLee/deep-learning-from-scratch-2 The above figure shows the structure of ____________ model and read line indicates the backpropagation flow. This model has an input, hidden, and output layer. The hidden layer contains a memory cell with three gates: input gate, forget gate, and output gate. By adding gate structures in a memory cell, this model can consider long-term dependencies from long-sequence data.

The structure of _________ model for the image classification is: Ref: https://github.com/WegraLee/deep-learning-from-scratch?tab=readme-ov-file An affine layer (i.e., a fully connected layer) means each input node in a layer is connected to all output nodes in the next layer. Ref: https://ml4a.github.io/ml4a/neural_networks As can be seen in the above figure, the deep learning model can classify the image dataset from zero to nine. First, input image data is converted to a 2-dimensional matrix (28 rows x 28 columns = 768 pixels). Second, the 2-dimensional matrix is converted into a vector for 768-pixels. If the first fully connected layer (i.e., affine layer) has 100 hidden nodes, there will be 76,800 connections. In this case, 76,800 parameters for weights and 76,800 parameters for constant terms are required to be estimated during the training step. As a result, this deep learning model requires high computational resources and time.

(1)________________(a. pre-training b. fine-tuning, c. hyperparameter-tuning d. overfitting; 5 points): using a pre-trained language model with a large corpus and does additional training with your dataset for a specific task. It requires relatively low computation and data resources but generally performs well. (2)________________(a. pre-training b. fine-tuning, c. hyperparameter-tuning d. overfitting; 5 points): is training the model with a large corpus from scratch; however, this approach requires high computational resources and big corpus data.