 Award ID(s):
 1911053
 NSFPAR ID:
 10375771
 Date Published:
 Journal Name:
 IEEE Transactions on Games
 ISSN:
 24751502
 Page Range / eLocation ID:
 1 to 1
 Format(s):
 Medium: X
 Sponsoring Org:
 National Science Foundation
More Like this

Obeid, I. ; Selesnik, I. ; Picone, J. (Ed.)The Neuronix highperformance computing cluster allows us to conduct extensive machine learning experiments on big data [1]. This heterogeneous cluster uses innovative scheduling technology, Slurm [2], that manages a network of CPUs and graphics processing units (GPUs). The GPU farm consists of a variety of processors ranging from lowend consumer grade devices such as the Nvidia GTX 970 to higherend devices such as the GeForce RTX 2080. These GPUs are essential to our research since they allow extremely computeintensive deep learning tasks to be executed on massive data resources such as the TUH EEG Corpus [2]. We use TensorFlow [3] as the core machine learning library for our deep learning systems, and routinely employ multiple GPUs to accelerate the training process. Reproducible results are essential to machine learning research. Reproducibility in this context means the ability to replicate an existing experiment – performance metrics such as error rates should be identical and floatingpoint calculations should match closely. Three examples of ways we typically expect an experiment to be replicable are: (1) The same job run on the same processor should produce the same results each time it is run. (2) A job run on a CPU and GPU should produce identical results. (3) A job should produce comparable results if the data is presented in a different order. System optimization requires an ability to directly compare error rates for algorithms evaluated under comparable operating conditions. However, it is a difficult task to exactly reproduce the results for large, complex deep learning systems that often require more than a trillion calculations per experiment [5]. This is a fairly wellknown issue and one we will explore in this poster. Researchers must be able to replicate results on a specific data set to establish the integrity of an implementation. They can then use that implementation as a baseline for comparison purposes. A lack of reproducibility makes it very difficult to debug algorithms and validate changes to the system. Equally important, since many results in deep learning research are dependent on the order in which the system is exposed to the data, the specific processors used, and even the order in which those processors are accessed, it becomes a challenging problem to compare two algorithms since each system must be individually optimized for a specific data set or processor. This is extremely timeconsuming for algorithm research in which a single run often taxes a computing environment to its limits. Wellknown techniques such as crossvalidation [5,6] can be used to mitigate these effects, but this is also computationally expensive. These issues are further compounded by the fact that most deep learning algorithms are susceptible to the way computational noise propagates through the system. GPUs are particularly notorious for this because, in a clustered environment, it becomes more difficult to control which processors are used at various points in time. Another equally frustrating issue is that upgrades to the deep learning package, such as the transition from TensorFlow v1.9 to v1.13, can also result in large fluctuations in error rates when rerunning the same experiment. Since TensorFlow is constantly updating functions to support GPU use, maintaining an historical archive of experimental results that can be used to calibrate algorithm research is quite a challenge. This makes it very difficult to optimize the system or select the best configurations. The overall impact of all of these issues described above is significant as error rates can fluctuate by as much as 25% due to these types of computational issues. Crossvalidation is one technique used to mitigate this, but that is expensive since you need to do multiple runs over the data, which further taxes a computing infrastructure already running at max capacity. GPUs are preferred when training a large network since these systems train at least two orders of magnitude faster than CPUs [7]. Largescale experiments are simply not feasible without using GPUs. However, there is a tradeoff to gain this performance. Since all our GPUs use the NVIDIA CUDA® Deep Neural Network library (cuDNN) [8], a GPUaccelerated library of primitives for deep neural networks, it adds an element of randomness into the experiment. When a GPU is used to train a network in TensorFlow, it automatically searches for a cuDNN implementation. NVIDIA’s cuDNN implementation provides algorithms that increase the performance and help the model train quicker, but they are nondeterministic algorithms [9,10]. Since our networks have many complex layers, there is no easy way to avoid this randomness. Instead of comparing each epoch, we compare the average performance of the experiment because it gives us a hint of how our model is performing per experiment, and if the changes we make are efficient. In this poster, we will discuss a variety of issues related to reproducibility and introduce ways we mitigate these effects. For example, TensorFlow uses a random number generator (RNG) which is not seeded by default. TensorFlow determines the initialization point and how certain functions execute using the RNG. The solution for this is seeding all the necessary components before training the model. This forces TensorFlow to use the same initialization point and sets how certain layers work (e.g., dropout layers). However, seeding all the RNGs will not guarantee a controlled experiment. Other variables can affect the outcome of the experiment such as training using GPUs, allowing multithreading on CPUs, using certain layers, etc. To mitigate our problems with reproducibility, we first make sure that the data is processed in the same order during training. Therefore, we save the data from the last experiment and to make sure the newer experiment follows the same order. If we allow the data to be shuffled, it can affect the performance due to how the model was exposed to the data. We also specify the float data type to be 32bit since Python defaults to 64bit. We try to avoid using 64bit precision because the numbers produced by a GPU can vary significantly depending on the GPU architecture [1113]. Controlling precision somewhat reduces differences due to computational noise even though technically it increases the amount of computational noise. We are currently developing more advanced techniques for preserving the efficiency of our training process while also maintaining the ability to reproduce models. In our poster presentation we will demonstrate these issues using some novel visualization tools, present several examples of the extent to which these issues influence research results on electroencephalography (EEG) and digital pathology experiments and introduce new ways to manage such computational issues.more » « less

null (Ed.)One of the most fundamental elements of narrative is character: if we are to understand a narrative, we must be able to identify the characters of that narrative. Therefore, character identification is a critical task in narrative natural language understanding. Most prior work has lacked a narratologically grounded definition of character, instead relying on simplified or implicit definitions that do not capture essential distinctions between characters and other referents in narratives. In prior work we proposed a preliminary definition of character that was based in clear narratological principles: a character is an animate entity that is important to the plot. Here we flesh out this concept, demonstrate that it can be reliably annotated (0.78 Cohen’s κ), and provide annotations of 170 narrative texts, drawn from 3 different corpora, containing 1,347 character coreference chains and 21,999 noncharacter chains that include 3,937 animate chains. Furthermore, we have shown that a supervised classifier using a simple set of easily computable features can effectively identify these characters (overall F1 of 0.90). A detailed error analysis shows that character identification is first and foremost affected by coreference quality, and further, that the shorter a chain is the harder it is to effectively identify as a character. We release our code and data for the benefit of other researchersmore » « less

Creating engaging interactive storybased experiences dynamically responding to individual player choices poses significant challenges for narrativecentered games. Recent advances in pretrained large language models (LLMs) have the potential to revolutionize procedural content generation for narrativecentered games. Historically, interactive narrative generation has specified pivotal events in the storyline, often utilizing planningbased approaches toward achieving narrative coherence and maintaining the story arc. However, manual authorship is typically used to create detail and variety in nonplayer character (NPC) interaction to specify and instantiate plot events. This paper proposes SCENECRAFT, a narrative scene generation framework that automates NPC interaction crucial to unfolding plot events. SCENECRAFT interprets natural language instructions about scene objectives, NPC traits, location, and narrative variations. It then employs large language models to generate game scenes aligned with authorial intent. It generates branching conversation paths that adapt to player choices while adhering to the author’s interaction goals. LLMs generate interaction scripts, semantically extract character emotions and gestures to align with the script, and convert dialogues into a game scripting language. The generated script can then be played utilizing an existing narrativecentered game framework. Through empirical evaluation using automated and human assessments, we demonstrate SCENECRAFT’s effectiveness in creating narrative experiences based on creativity, adaptability, and alignment with intended author instructions.

Jin, Shi (Ed.)In this paper, we apply the idea of fictitious play to design deep neural networks (DNNs), and develop deep learning theory and algorithms for computing the Nash equilibrium of asymmetric Nplayer nonzerosum stochastic differential games, for which we refer as deep fictitious play, a multistage learning process. Specifically at each stage, we propose the strategy of letting individual player optimize her own payoff subject to the other players’ previous actions, equivalent to solving N decoupled stochastic control optimization problems, which are approximated by DNNs. Therefore, the fictitious play strategy leads to a structure consisting of N DNNs, which only communicate at the end of each stage. The resulting deep learning algorithm based on fictitious play is scalable, parallel and modelfree, i.e., using GPU parallelization, it can be applied to any Nplayer stochastic differential game with different symmetries and heterogeneities (e.g., existence of major players). We illustrate the performance of the deep learning algorithm by comparing to the closedform solution of the linear quadratic game. Moreover, we prove the convergence of fictitious play under appropriate assumptions, and verify that the convergent limit forms an openloop Nash equilibrium. We also discuss the extensions to other strategies designed upon fictitious play and closedloop Nash equilibrium in the end.more » « less

Expander graphs play a central role in graph theory and algorithms. With a number of powerful algorithmic tools developed around them, such as the CutMatching game, expander pruning, expander decomposition, and algorithms for decremental AllPairs Shortest Paths (APSP) in expanders, to name just a few, the use of expanders in the design of graph algorithms has become ubiquitous. Specific applications of interest to us are fast deterministic algorithms for cut problems in static graphs, and algorithms for dynamic distancebased graph problems, such as APSP. Unfortunately, the use of expanders in these settings incurs a number of drawbacks. For example, the best currently known algorithm for decremental APSP in constantdegree expanders can only achieve a (log n) O(1/ 2 ) approximation with n 1+O( ) total update time for any . All currently known algorithms for the Cut Player in the CutMatching game are either randomized, or provide rather weak guarantees: expansion 1/(log n) 1/ with running time n 1+O( ) . This, in turn, leads to somewhat weak algorithmic guarantees for several central cut problems: the best current almost linear time deterministic algorithms for Sparsest Cut, Lowest Conductance Cut, and Balanced Cut can only achieve approximation factor (log n) ω(1). Lastly, when relying on expanders in distancebased problems, such as dynamic APSP, via current methods, it seems inevitable that one has to settle for approximation factors that are at least Ω(log n). In contrast, we do not have any negative results that rule out a factor5 approximation with nearlinear total update time. In this paper we propose the use of wellconnected graphs, and introduce a new algorithmic toolkit for such graphs that, in a sense, mirrors the above mentioned algorithmic tools for expanders. One of these new tools is the Distanced Matching game, an analogue of the CutMatching game for wellconnected graphs. We demonstrate the power of these new tools by obtaining better results for several of the problems mentioned above. First, we design an algorithm for decremental APSP in expanders with significantly better guarantees: in a constantdegree expander, the algorithm achieves (log n) 1+o(1)approximation, with total update time n 1+o(1). We also obtain a deterministic algorithm for the Cut Player in the CutMatching game that achieves expansion 1 (log n) 5+o(1) in time n 1+o(1), deterministic almost lineartime algorithms for Sparsest Cut, LowestConductance Cut, and Minimum Balanced Cut with approximation factors O(poly log n), as well as improved deterministic algorithm for Expander Decomposition. We believe that the use of wellconnected graphs instead of expanders in various dynamic distancebased problems (such as APSP in general graphs) has the potential of providing much stronger guarantees, since we are no longer necessarily restricted to superlogarithmic approximation factors.more » « less