Applying machine learning techniques to solve production problems within electronic design automation is complex. This is because production engineering applications have accuracy, scalability, complexity, verifiability, and usability requirements that are not met by traditional machine learning approaches. These additional challenges are often not well understood or adequately solved in practice, which causes production machine learning approaches to fail. This invited paper examines these engineering-specific challenges and presents some effective solutions based on Solido's experience developing a suite of successful applied machine learning solutions for EDA over the past twelve years.

Real-time distributed groupware sends several kinds of messages with varying quality-of-service requirements. However, standard network protocols do not provide the flexibility needed to support these different requirements (either providing too much reliability or too little), leading to poor performance on real-world networks. To address this problem, we investigated the use of an application-level networking technique called adaptive forward error correction (AFEC) for real-time groupware. AFEC can maintain a predefined level of reliability while avoiding the overhead of packet acknowledgement or retransmission. We analysed the requirements of typical real-time groupware systems and developed an AFEC technique to meet these needs. We tested the new technique in an experiment that measured message reliability and latency using TCP, plain UDP, UDP with non-adaptive FEC, and UDP with our AFEC scheme, under several simulated network conditions. Our results show that for awareness messages that can tolerate some loss, FEC approaches keep latency at nearly the plain-UDP level while dramatically improving reliability. In addition, adaptive FEC is the only technique that can maintain a specified level of reliability and also minimize delay as network conditions change. Our study shows that groupware AFEC can be a useful tool for improving the real-world performance and usability of real-time groupware.

Networked games can provide groupware developers with important lessons in how to deal with real-world networking issues such as latency, limited bandwidth and packet loss. Games have similar demands and characteristics to groupware, but unlike the applications studied by academics, games have provided production-quality real-time interaction for many years. The techniques used by games have not traditionally been made public, but several game networking libraries have recently been released as open source, providing the opportunity to learn how games achieve network performance. We examined five game libraries to find networking techniques that could benefit groupware; this paper presents the concepts most valuable to groupware developers, including techniques to deal with limited bandwidth, reliability, and latency. Some of the techniques have been previously reported in the networking literature; therefore, the contribution of this paper is to survey which techniques have been shown to work, over several years, and then to link these techniques to quality requirements specific to groupware. By adopting these techniques, groupware designers can dramatically improve network performance on the real-world Internet.

Groupware communicates by sending messages across the network, and groupware programmers use a variety of formats for these messages, such as XML, plain text, or serialized objects. Although these formats have many advantages, they are often so verbose that they overload the system's network resources. Groupware programmers could improve efficiency by using more compact formats, but this efficiency comes at the cost of increased complexity, reduced convenience, and reduced readability. In this paper we propose an alternate approach for improving efficiency -- an automatic compression system that transparently minimizes verbose formats. Our general message compressor -- GMC -- automatically finds and removes redundancy in message streams, without any knowledge of the contents or structure of the message, and without any need for the programmer to change the way they work. In tests with realistic message traces, GMC reduced text messages to 20% of their original size, XML messages to 8% of the original, and serialized objects to 9%. Although not as compact as a hand-coded representation, GMC provides most of the compression benefits with almost none of the work -- it allows groupware programmers to use convenient message formats without compromising transport efficiency.

Delay is an unavoidable reality in collaborative environments. We propose an approach to dealing with delay in which 'decorators' are introduced into the interface. Decorators show the presence, magnitude and effects of delay so that participants can better understand its consequences and adopt their own natural coping strategies. Two experiments with different decorators show that this approach can significantly reduce errors in specific collaborative activities. We conclude that revealing delays is one way in which groupware can benefit from accepting and working with the reality of distributed systems, rather than trying to maintain the illusion of copresent interaction.

Although telepointers are valuable for supporting real-time collaboration, they are rarely seen in commercial groupware applications that run on the Internet. One reason for their absence is that current telepointer implementations perform poorly on real-world networks with varying traffic, congestion, and loss. In this paper, we report on a new implementation of telepointers (HPT) that is designed to provide smooth, timely, and accurate telepointers in real-world groupware: on busy networks, on cable and dialup connections, and on wireless channels. HPT maintains performance at usable levels with a combination of techniques from multimedia and distributed systems research, including UDP transport, message compression, motion prediction, adaptive rate control, and adaptive forward error correction. Although these techniques have been seen before, they have never been combined and tailored to the specific requirements of telepointers. Tests of the new implementation show that HPT provides good performance in a number of network situations where other implementations do not work at all - we can provide usable telepointers even over a lossy 28K modem connection. HPT sets a new standard for telepointers, and allows designers to greatly improve the support that groupware provides for real-time interaction over distance.

Telepointers are an important type of embodiment in real-time distributed groupware. Telepointers can increase the presence of remote participants and can provide considerable awareness information about people's locations and activities. However, the motion of a telepointer is often disrupted by network jitter. Although some strategies exist for dealing with jitter, none of these techniques are able to restore the immediacy and smoothness of a real cursor. In this paper we investigate the use of prediction - commonly used in networked simulations and games - to reduce the effects of jitter on telepointer motion. To determine whether prediction can be effective for improving telepointers, we carried out two experiments that tested the effects of different prediction schemes (some real and some artificial) on people's ability to interpret telepointer gestures. These studies show that although cursor prediction is still a difficult problem, there are both potential performance improvements, and definite preference advantages. Our studies suggest that telepointer prediction should be routinely used to increase the immediacy and naturalness of remote interaction, and suggest that prediction can also improve interpretation in certain situations.