Abstract: Chemistry42 is a software platform for de novo small molecule design that integrates Artificial Intelligence (AI) techniques with computational and medicinal chemistry methods. Chemistry42 is unique in its ability to generate novel molecular structures with predefined properties validated through in vitro and in vivo studies. Keywords: generative chemistry, target identification, deep learning, reinforcement learning, drug discovery, de novo drug design Introduction Deep Learning (DL) has proven to be very effective in speech and image recognition. This is because DL-based architectures are uniquely suited for the automatic identification of patterns within complex, nonlinear data sets without the need for manual feature engineering. DL methods have successfully overcome limitations inherent in the standard techniques used for small molecule design (Chen et al. 2018; Vanhaelen, Lin, and Zhavoronkov 2020; Yang et al. 2019) which offers exciting possibilities for the development of new methods that efficiently explore uncharted chemical space.

Twitter is a microblogging and social media platform that allows users to communicate via short messages (280 characters) known as tweets [1, 2, 3]. It enables millions of users to express their opinions on a daily basis on a variety of different topics ranging from reviews on products and services to users' political and religious views, making Twitter a potent tool for gauging public sentiment [4]. Thus, it manifestly follows that twitter data can be regarded as a corpus, forming the basis on which predictions can be made, and researchers have indeed exploited this fact to seek trends by performing numerous and varied analyses. A characteristic feature of the stock market is volatility and there is no general equation describing the prediction of stock prices, which is a complex function of a range of different factors. The methods of stock market prediction can be broadly classified into Technical Analysis and Fundamental Analysis [5]. The latter involves the consideration of macroeconomic factors as well as industry specific news and events to guide investment strategies [5]. The analysis of public sentiment via tweets performed in this project can be regarded as an aspect of Fundamental Analysis. Although the prediction of stock prices is highly nuanced, the Efficient Market Hypothesis (EMH), propounded by Eugene Farma in the 1960's, suggested a relation between public opinion and stock prices [6]. The semi-strong form of the EMH implies that current events and new public information have a significant bearing on market trends [1, 6].

In a low-rise building overlooking a busy intersection in Beijing, Ji Rong Wen, a middle-aged scientist with thin-rimmed glasses and a mop of black hair, excitedly describes a project that could advance one of the hottest areas of artificial intelligence. Wen leads a team at the Beijing Academy of Artificial Intelligence (BAAI), a government-sponsored research lab that's testing a powerful new language algorithm--something similar to GPT-3, a program revealed in June by researchers at OpenAI that digests large amounts of text and can generate remarkably coherent, free-flowing language. "This is a big project," Wen says with a big grin. "It takes a lot of computing infrastructure and money." Wen, a professor at Renmin University in Beijing recruited to work part-time at BAAI, hopes to create an algorithm that is even cleverer than GPT-3. He plans to combine machine learning with databases of facts, and to feed the algorithm images and video as well as text, in hope of creating a richer understanding of the physical world--that the words cat and fur don't just often appear in the same sentence, but are associated with one another visually.

Event-driven sensors such as LiDAR and dynamic vision sensor (DVS) have found increased attention in high-resolution and high-speed applications. A lot of work has been conducted to enhance recognition accuracy. However, the essential topic of recognition delay or time efficiency is largely under-explored. In this paper, we present a spiking learning system that uses the spiking neural network (SNN) with a novel temporal coding for accurate and fast object recognition. The proposed temporal coding scheme maps each event's arrival time and data into SNN spike time so that asynchronously-arrived events are processed immediately without delay. The scheme is integrated nicely with the SNN's asynchronous processing capability to enhance time efficiency. A key advantage over existing systems is that the event accumulation time for each recognition task is determined automatically by the system rather than pre-set by the user. The system can finish recognition early without waiting for all the input events. Extensive experiments were conducted over a list of 7 LiDAR and DVS datasets. The results demonstrated that the proposed system had state-of-the-art recognition accuracy while achieving remarkable time efficiency. Recognition delay was shown to reduce by 56.3% to 91.7% in various experiment settings over the popular KITTI dataset.

For a d-ary Boolean function Φ: {0, 1}d → {0, 1} and an assignment to its variables x = (x1, x2, . . . , xd) we consider the problem of finding those subsets of the variables that are sufficient to determine the function value with a given probability δ. This is motivated by the task of interpreting predictions of binary classifiers described as Boolean circuits, which can be seen as special cases of neural networks. We show that the problem of deciding whether such subsets of relevant variables of limited size k ≤ d exist is complete for the complexity class NPPP and thus, generally, unfeasible to solve. We then introduce a variant, in which it suffices to check whether a subset determines the function value with probability at least δ or at most δ − γ for 0 < γ < δ. This promise of a probability gap reduces the complexity to the class NPBPP. Finally, we show that finding the minimal set of relevant variables cannot be reasonably approximated, i.e. with an approximation factor d1−α for α > 0, by a polynomial time algorithm unless P = NP. This holds even with the promise of a probability gap.

Action recognition, early prediction, and online action detection are complementary disciplines that are often studied independently. Most online action detection networks use a pre-trained feature extractor, which might not be optimal for its new task. We address the task-specific feature extraction with a teacher-student framework between the aforementioned disciplines, and a novel training strategy. Our network, Online Knowledge Distillation Action Detection network (OKDAD), embeds online early prediction and online temporal segment proposal subnetworks in parallel. Low interclass and high intraclass similarity are encouraged during teacher training. Knowledge distillation to the OKDAD network is ensured via layer reuse and cosine similarity between teacher-student feature vectors. Layer reuse and similarity learning significantly improve our baseline which uses a generic feature extractor. We evaluate our framework on infrared videos from two popular datasets, NTU RGB+D (action recognition, early prediction) and PKU MMD (action detection). Unlike previous attempts on those datasets, our student networks perform without any knowledge of the future. Even with this added difficulty, we achieve state-of-the-art results on both datasets. Moreover, our networks use infrared from RGB-D cameras, which we are the first to use for online action detection, to our knowledge.

Deep learning has demonstrated its strengths in numerous binary analysis tasks, including function boundary detection, binary code search, function prototype inference, value set analysis, etc. When applying deep learning to binary analysis tasks, we need to decide what input should be fed into the neural network model. More specifically, we need to answer how to represent an instruction in a fixed-length vector. The idea of automatically learning instruction representations is intriguing, however the existing schemes fail to capture the unique characteristics of disassembly. These schemes ignore the complex intra-instruction structures and mainly rely on control flow in which the contextual information is noisy and can be influenced by compiler optimizations. In this paper, we propose to pre-train an assembly language model called PalmTree for generating general-purpose instruction embeddings by conducting self-supervised training on large-scale unlabeled binary corpora. PalmTree utilizes three pre-training tasks to capture various characteristics of assembly language. These training tasks overcome the problems in existing schemes, thus can help to generate high-quality representations. We conduct both intrinsic and extrinsic evaluations, and compare PalmTree with other instruction embedding schemes. PalmTree has the best performance for intrinsic metrics, and outperforms the other instruction embedding schemes for all downstream tasks.

Geometry of neural network loss landscape has been studied via the analysis of the global optimality of local minima [1, 2, 5, 8, 9, 14], the existence of a continuous descending path to a global optimum [4, 7, 10, 11, 13], the connectivity of dropout-stable solutions [6, 12], and the topology of sublevel sets [7, 13]. In this paper, we improve the result of [7]. In particular, [7] shows that for a general class of convex loss functions (e.g.

The need for analysis of toxicity in new drug candidates and the requirement of doing it fast have asked the consideration of scientists towards the use of artificial intelligence tools to examine toxicity levels and to develop models to a degree where they can be used commercially to measure toxicity levels efficiently in upcoming drugs. Artificial Intelligence based models can be used to predict the toxic nature of a chemical using Quantitative Structure Activity Relationship techniques. Convolutional Neural Network models have demonstrated great outcomes in predicting the qualitative analysis of chemicals in order to determine the toxicity. This paper goes for the study of Simplified Molecular Input Line-Entry System (SMILES) as a parameter to develop Long short term memory (LSTM) based models in order to examine the toxicity of a molecule and the degree to which the need can be fulfilled for practical use alongside its future outlooks for the purpose of real world applications.

Deep Learning (DL) is a machine learning procedure for artificial intelligence that analyzes the input data in detail by increasing neuron sizes and number of the hidden layers. DL has a popularity with the common improvements on the graphical processing unit capabilities. Increasing number of the neuron sizes at each layer and hidden layers is directly related to the computation time and training speed of the classifier models. The classification parameters including neuron weights, output weights, and biases need to be optimized for obtaining an optimum model. Most of the popular DL algorithms require long training times for optimization of the parameters with feature learning progresses and back-propagated training procedures. Reducing the training time and providing a real-time decision system are the basic focus points of the novel approaches. Deep Extreme Learning machines (Deep ELM) classifier model is one of the fastest and effective way to meet fast classification problems. In this study, Deep ELM model, its superiorities and weaknesses are discussed, the problems that are more suitable for the classifiers against Convolutional neural network based DL algorithms.