Created at 11pm, Apr 30
buaziziTechnology
0
Emerging Trend in Applied Deep Learning Research
Contract ID
cVLQzXCJWzw0QMhfUqKRVJ1mj0baPLvuAm8oYjsc16s
File Type
PDF
Entry Count
54
Embed. Model
jina_embeddings_v2_base_en
Index Type
hnsw

The recent emerging trending of artificial intelligent has drawn a significant attention inacademia and industry. Such trend should be given credit to the recent rise of deep learning,a specific type of machine learning method mimicking the neural network behavior in thehuman brain that achieves significant performance powered by modern super-hardware.Various empirical studies indicate the superiority of deep learning among other machinelearning algorithms. In this report, I will discuss the rising trend of applied field of deeplearning in the research community. Specifically, I will talk about three trending fields,transfer learning, deep reinforcement learning, and recommender system. The recent innovations as well as challenges and potential applications will be discussed. Although I discussstate-of-the-art research in this report, I target to make readers who have limited knowledgein artificial intelligence to understand the novel applications in this tremendously potentialyet inconclusive area.

(2016) proposed the residual neural network, an inuential work which enables researchers to train a network in depth of 100 layers, and received the best performance in various tasks, such as image recognition.
id: ec8617a1898ae8a6db025391c7fa34f1 - page: 9
In the recent two years (since 2018), natural language processing has experienced a big revolution by pre-training the language model, which results in human-level performance in tasks such as reading comprehension, sentiment analysis, machine translation. To represent words and paragraph of a corpus of texts, Mikolov et al. (2013) introduced a word2vector to annotate extraction of semantic representation of words by denoting each word into an independent vector. Such word vectors can be applied in other NLP tasks. However, the biggest challenge of such word representation is the inability to share representation at sub-word levels. In other words, if one word has different meanings (e.g. bank has both meaning of the land alongside to a river or lake or a place to store money), the representation of word2vector cannot distinguish such differences. Researchers try to explore the option of deep learning using an enormous text corpus in the internet. Peters et al. (2018) introduced a bi
id: 8ce767733c0d303e8b068bb3b3c895f9 - page: 9
-directional LSTM network to create a rich language representative model. Devlin et al. (2018) applied the bi -directional Transformer model training in an extremely large dataset, which leads to an impressive performance boosts.
id: 2ff68c479e3f1a48f9625d41cf405aa7 - page: 9
The future research will not be limited in the area merely in computer vision and natural language processing. For instance, Spruyt (2018) proposed a convolutional neural network framework to represent a geolocation. Such method can be used directly for tasks such as venue mapping or transport classication and contribute to improve the classier accuracies and generalization capabilities by means of transfer learning.
id: defcdb357dd888442463d428990b35bb - page: 9
How to Retrieve?
# Search

curl -X POST "https://search.dria.co/hnsw/search" \
-H "x-api-key: <YOUR_API_KEY>" \
-H "Content-Type: application/json" \
-d '{"rerank": true, "top_n": 10, "contract_id": "cVLQzXCJWzw0QMhfUqKRVJ1mj0baPLvuAm8oYjsc16s", "query": "What is alexanDRIA library?"}'
        
# Query

curl -X POST "https://search.dria.co/hnsw/query" \
-H "x-api-key: <YOUR_API_KEY>" \
-H "Content-Type: application/json" \
-d '{"vector": [0.123, 0.5236], "top_n": 10, "contract_id": "cVLQzXCJWzw0QMhfUqKRVJ1mj0baPLvuAm8oYjsc16s", "level": 2}'